Apparatus and method of determining a cardiac implant size and methods and systems for transesophageal echocardiogram guided implantation of left atrial appendage closure device

Abstract

Apparatus and methods of determining a cardiac implant size and methods and system for transesophageal echocardiogram guided implantation of left atrial appendage closure device. The apparatus includes an ultrasonic imaging device, at least a processor, and a memory communicatively connected to the at least a processor, wherein the memory contains instructions configuring the at least a processor to using the ultrasonic imaging device collect a plurality of ultrasonic images using a 3D cardiac model generation machine learning model trained on a training dataset including example ultrasonic images correlated with example 3D cardiac models. The processor is further configured to generate a 3D cardiac model based on the plurality of ultrasonic images, generate at least a cardiac measurement based on the 3D cardiac model, determine a cardiac implant size based on the at least a cardiac measurement, and using a display, display to a user the cardiac implant size.

Description

[0001] APPARATUS AND METHOD OF DETERMINING A CARDIAC IMPLANT SIZE AND METHODS AND SYSTEMS FOR TRANSESOPHAGEAL ECHOCARDIOGRAM GUIDED IMPLANTATION OF LEFT ATRIAL APPENDAGE CLOSURE DEVICE

[0002] REFERENCE TO RELATED APPLICATIONS

[0003] This application claims the benefit of priority of U.S. Nonprovisional Patent Application Serial No. 19 / 046,936, filed on February 6, 2025, and titled “METHODS AND SYSTEMS FOR TRANSESOPHAGEAL ECHOCARDIOGRAM GUIDED IMPLANTATION OF LEFT ATRIAL APPENDAGE CLOSURE DEVICE,” which claims the benefit of priority of U.S. Provisional Patent Application Serial No. 63 / 705,376, filed on October 9, 2024, and titled “METHODS AND SYSTEMS FOR TRANSESOPHAGEAL ECHOCARDIOGRAM GUIDED IMPLANTATION OF LEFT ATRIAL APPENDAGE CLOSURE DEVICE”, the application also claims the benefit of priority of U.S. Nonprovisional Patent Application Serial No. 18 / 938,980, filed on November 6, 2024, and titled “APPARATUS AND METHOD OF DETERMINING A CARDIAC IMPLANT SIZE”. Each applications is incorporated by reference herein in its entirety.

[0004] FIELD OF THE INVENTION

[0005] The present invention generally relates to the field of cardiac imaging and left atrial appendage closure. In particular, the present invention is directed to apparatus and methods of determining a cardiac implant size and methods and system for transesophageal echocardiogram guided implantation of left atrial appendage closure device.

[0006] BACKGROUND

[0007] Computed tomography (CT) has been shown to be superior to 2D TEE in Watchman preprocedural device sizing. As per the PRO3DLAAO clinical trial, the accuracy for 1st occluder device selection is 92% with CT vs. 27% with 2D-TEE, using the final implanted Watchman device size as the reference standard. The poor accuracy of the initial device selection results in longer procedures and usage of multiple devices until the implantation can be verified to be leakage-free. That is, an average of 2.5 devices is used with 2D TEE-based planning vs. 1.3 devices used with CT-based planning. However, computed tomography exposes subjects to radiation, and is often performed on a separate date, requiring multiple appointments and potentially reducing accuracy due to the time delay between when the data is captured and when

[0008] 1 Attorney Docket No. 1518-193PCT1 it is used.

[0009] Atrial fibrillation (AF) is one of the most common cardiac arrhythmias, affecting millions of individuals worldwide. AF significantly increases the risk of thromboembolic events, particularly ischemic stroke, due to the formation of blood clots in the left atrial appendage (LAA), a small pouch located in the left atrium of the heart. The irregular heart rhythms associated with AF can lead to blood pooling in the LAA, creating conditions conducive to clot formation. For patients with non-valvular atrial fibrillation (NVAF), the LAA is estimated to be the source of thrombi in the majority of ischemic strokes. Left atrial appendage closure devices must be implanted into a patient’s heart to close a left atrial appendage.

[0010] SUMMARY OF THE DISCLOSURE

[0011] In an aspect, described herein is an apparatus for determining a cardiac implant size. Such an apparatus may include an ultrasonic imaging device, at least a processor, and a memory communicatively connected to the at least a processor, wherein the memory contains instructions configuring the at least a processor to, using the ultrasonic imaging device, collect a plurality of ultrasonic images, using a 3D cardiac model generation machine learning model trained on a training dataset comprising example ultrasonic images correlated with example 3D cardiac models, generate a 3D cardiac model based on the plurality of ultrasonic images, generate at least a cardiac measurement based on the 3D cardiac model, determine a cardiac implant size based on the at least a cardiac measurement, and display to a user the cardiac implant size.

[0012] In one or more embodiments, a cardiac implant size may include a left atrial appendage occlusion device size.

[0013] In one or more embodiments, a memory may include instructions further configuring a processor to determine one or both of a cardiac implant placement and a cardiac implant candidate as a function of a 3D cardiac model.

[0014] In one or more embodiments, a memory may include instructions further configuring a processor to receive at least an ultrasound localization datum representing one or both of position and angle of an ultrasonic imaging device, and generate, using a 3D cardiac model generation machine learning model, a 3D cardiac model as a function of the at least an ultrasound localization datum.

[0015] In one or more embodiments, a plurality of ultrasonic images may include one or more of transesophageal echocardiogram (TEE) images, intracardiac echocardiogram (ICE) images,

[0016] 2 Attorney Docket No. 1518-193PCT1 transthoracic echocardiogram (TTE) images, and point of care ultrasound (POCUS) images.

[0017] In one or more embodiments, a memory may include instructions further configuring a processor to calculate a level of uncertainty at a plurality of locations on a 3D cardiac model, wherein a plurality of locations may include a high uncertainty region, receive a subsequent plurality of ultrasonic images of cardiac anatomy corresponding to a high uncertainty region of a 3D cardiac model, wherein a subsequent plurality of ultrasonic images is captured using an ultrasonic imaging device as a function of a high uncertainty region, and generate a subsequent 3D cardiac model as a function of a subsequent plurality of ultrasonic images.

[0018] In one or more embodiments, generating a 3D cardiac model based on a plurality of ultrasonic images may include generating a set of shape parameters representing a cardiac shape as a function of a plurality of ultrasonic images and a shape identification model trained on a training dataset, wherein a set of shape parameters may include a plurality of numerical descriptors representing at least a geometric characteristic of a subject’s heart and a plurality of associated parameter ranges.

[0019] In one or more embodiments, a memory may include instructions further configuring a processor to receive a 3D cardiac implant model representing a cardiac implant and display, using a display, a 3D cardiac implant model positioned within a 3D cardiac model.

[0020] In one or more embodiments, a cardiac measurement may include an ostial diameter of a left atrial appendage (LAA).

[0021] In one or more embodiments, generating a 3D cardiac model based on a plurality of ultrasonic images may include generating a 3D voxel occupancy representation (VOR) representing a cardiac shape as a function of a plurality of ultrasonic images and a 3D cardiac model generation machine learning model trained on a training dataset, generating a mesh representing a cardiac shape as a function of a 3D voxel occupancy representation, and displaying, using a display, a mesh to a user.

[0022] In another aspect, described herein is a method of determining a cardiac implant size, the method comprising, using at least a processor and an ultrasonic imaging device, collecting a plurality of ultrasonic images, using the at least a processor and a 3D cardiac model generation machine learning model trained on a training dataset comprising example ultrasonic images correlated with example 3D cardiac models, generating a 3D cardiac model based on the plurality of ultrasonic images, using the at least a processor, generating at least a cardiac

[0023] 3 Attorney Docket No. 1518-193PCT1 measurement based on the 3D cardiac model, using the at least a processor, determining a cardiac implant size based on the at least a cardiac measurement, and using the at least a processor, displaying to a user the cardiac implant size.

[0024] In one or more embodiments, a cardiac implant size may include a left atrial appendage occlusion device size.

[0025] In one or more embodiments, a method may further include determining one or both of a cardiac implant placement and a cardiac implant candidate quality as a function of a 3D cardiac model.

[0026] In one or more embodiments, a method may further include receiving, using at least a processor, at least an ultrasound localization datum representing one or both of position and angle of an ultrasonic imaging device, and generating, using at least a processor and a 3D cardiac model generation machine learning model, a 3D cardiac model as a function of the at least an ultrasound localization datum.

[0027] In one or more embodiments, a plurality of ultrasonic images may include one or more of transesophageal echocardiogram (TEE) images, intracardiac echocardiogram (ICE) images, transthoracic echocardiogram (TTE) images, and point of care ultrasound (POCUS) images.

[0028] In one or more embodiments, a method may further include calculating, using at least a processor, a level of uncertainty at a plurality of locations on a 3D cardiac model, wherein a plurality of locations may include a high uncertainty region, receiving, using at least a processor, a subsequent plurality of ultrasonic images of cardiac anatomy corresponding to a high uncertainty region of a 3D cardiac model, wherein a subsequent plurality of ultrasonic images is captured using an ultrasonic imaging device as a function of a high uncertainty region, and generating, using at least a processor, a subsequent 3D cardiac model as a function of a subsequent plurality of ultrasonic images.

[0029] In one or more embodiments, generating a 3D cardiac model based on a plurality of ultrasonic images may include generating a set of shape parameters representing a cardiac shape as a function of a plurality of ultrasonic images and a shape identification model trained on a training dataset, wherein a set of shape parameters may include a plurality of numerical descriptors representing at least a geometric characteristic of a subject’s heart and a plurality of associated parameter ranges.

[0030] In one or more embodiments, a method may further include receiving, using at least a

[0031] 4 Attorney Docket No. 1518-193PCT1 processor, a 3D cardiac implant model representing a cardiac implant and displaying, using at least a processor and a display, a 3D cardiac implant model positioned within a 3D cardiac model.

[0032] In one or more embodiments, a cardiac measurement may include an ostial diameter of a left atrial appendage (LAA).

[0033] In one or more embodiments, generating a 3D cardiac model based on a plurality of ultrasonic images may include generating, using at least a processor, a 3D voxel occupancy representation (VOR) representing a cardiac shape as a function of a plurality of ultrasonic images and a 3D cardiac model generation machine learning model trained on a training dataset, generating, using at least a processor, a mesh representing a cardiac shape as a function of a 3D voxel occupancy representation, and displaying, using at least a processor and a display, a mesh to a user.

[0034] In an aspect, a system for transesophageal echocardiogram-guided implantation of a left atrial appendage closure device is disclosed. The system includes at least a transesophageal echocardiogram (TEE) system including at least an ultrasound sensor, wherein the at least an ultrasound sensor is configured to be located within an esophagus of a patient and detect at least an ultrasound image as a function of cardiac tissue of the patient, at least a display and at least a computing device including at least a processor and a memory containing instructions configuring the at least a processor to receive the at least an ultrasound image, generate at least a three-dimensional (3D) cardiac model representative of a heart of the patient as a function of the at least an ultrasound image, receive at least a 3D left atrial appendage closure (LAAC) device model representative of at least a LAAC device, generate a superimposed model by superimposing the at least a 3D LAAC device model onto the at least a 3D cardiac model and display, using the at least a display, the superimposed model.

[0035] In one or more embodiments, receiving at least an ultrasound image may include extracting at least a TEE angle datum from at least an ultrasound image using optical character recognition, and generating an image inquiry datum as a function of at least a TEE angle datum and device instruction for use (IFU) data retrieved from an LAAC database, wherein an image inquiry datum may be configured to query additional ultrasound images to a user of at least a TEE system through at least a display.

[0036] In one or more embodiments, receiving at least an ultrasound image may include

[0037] 5 Attorney Docket No. 1518-193PCT1 generating view training data, wherein view training data may include exemplary ultrasound images correlated to exemplary view labels, training a view classifier using view training data, classifying at least an ultrasound image to at least a view label using a trained view classifier, and generating an image inquiry datum as a function of a view label and IFU data.

[0038] In one or more embodiments, generating at least a 3D cardiac model may include extracting at least a cardiac feature from at least an ultrasound image and segmenting at least an ultrasound image into a plurality of image segments.

[0039] In one or more embodiments, generating at least a 3D cardiac model may include generating a 3D point cloud as a function of a plurality of image segments and generating a 3D mesh model of at least a 3D cardiac model as a function of a 3D point cloud.

[0040] In one or more embodiments, generating at least a 3D cardiac model may include generating at least a 3D cardiac model using a statistical shape model.

[0041] In one or more embodiments, receiving at least a 3D LAAC device model may include extracting at least an ostium characteristic datum from at least an ultrasound image, determining a device datum as a function of at least an ostium characteristic datum and a compression rate of a plurality of LAAC devices, wherein a device datum may include a size datum, and generating at least a 3D LAAC device model as a function of a device datum.

[0042] In one or more embodiments, determining a device datum may include simulating a placement of a plurality of LAAC devices within at least a 3D cardiac model as a function of at least an ostium characteristic datum and a compression rate, and generating a device datum as a function of a simulation.

[0043] In one or more embodiments, determining a device datum may include determining a pass datum as a function of at least an ostium characteristic datum and a compression rate.

[0044] In one or more embodiments, superimposing at least a 3D LAAC device model onto at least a 3D cardiac model may include determining an optimal path for a placement of at least a 3D LAAC device model within at least a 3D cardiac model, generating a path model for an optimal path, and superimposing a path model onto at least a 3D cardiac model.

[0045] In one or more embodiments, a 3D cardiac model may include peripheral vasculature.

[0046] In one or more embodiments, generating at least a 3D cardiac model may include generating at least a 3D cardiac model using a point completion model.

[0047] In another aspect, a method for transesophageal echocardiogram-guided implantation of a

[0048] 6 Attorney Docket No. 1518-193PCT1 left atrial appendage closure device is disclosed. The method includes receiving, using at least a processor of a computing device, at least an ultrasound image from at least a transesophageal echocardiogram (TEE) system including at least an ultrasound sensor, wherein the at least an ultrasound sensor is configured to be located within an esophagus of a patient and detect the at least an ultrasound image as a function of cardiac tissue of the patient, generating, using the at least a processor, at least a three-dimensional (3D) cardiac model representative of a heart of the patient as a function of the at least an ultrasound image, receiving, using the at least a processor, at least a 3D left atrial appendage closure (LAAC) device model representative of at least a LAAC device, generating, using the at least a processor, a superimposed model by superimposing the at least a 3D LAAC device model onto the at least a 3D cardiac model and displaying, using the at least a processor, and at least a display, the superimposed model.

[0049] In one or more embodiments, receiving at least an ultrasound image may include extracting at least a TEE angle datum from at least an ultrasound image using optical character recognition, and generating an image inquiry datum as a function of at least a TEE angle datum and device instruction for use (IFU) data retrieved from an LAAC database, wherein an image inquiry datum may be configured to query additional ultrasound images to a user of at least a TEE system through at least a display.

[0050] In one or more embodiments, receiving at least an ultrasound image may include generating view training data, wherein view training data may include exemplary ultrasound images correlated to exemplary view labels, training a view classifier using view training data, classifying at least an ultrasound image to at least a view label using a trained view classifier, and generating an image inquiry datum as a function of a view label and IFU data.

[0051] In one or more embodiments, generating at least a 3D cardiac model may include extracting at least a cardiac feature from at least an ultrasound image and segmenting at least an ultrasound image into a plurality of image segments.

[0052] In one or more embodiments, generating at least a 3D cardiac model may include generating a 3D point cloud as a function of a plurality of image segments and generating a 3D mesh model of at least a 3D cardiac model as a function of a 3D point cloud.

[0053] In one or more embodiments, generating at least a 3D cardiac model may include generating at least a 3D cardiac model using a statistical shape model.

[0054] In one or more embodiments, receiving at least a 3D LAAC device model may include

[0055] 7 Attorney Docket No. 1518-193PCT1 extracting at least an ostium characteristic datum from at least an ultrasound image, determining a device datum as a function of at least an ostium characteristic datum and a compression rate of a plurality of LAAC devices, wherein a device datum may include a size datum, and generating at least a 3D LAAC device model as a function of a device datum.

[0056] In one or more embodiments, determining a device datum may include simulating a placement of a plurality of LAAC devices within at least a 3D cardiac model as a function of at least an ostium characteristic datum and a compression rate, and generating a device datum as a function of a simulation.

[0057] In one or more embodiments, determining a device datum may include determining a pass datum as a function of at least an ostium characteristic datum and a compression rate.

[0058] In one or more embodiments, superimposing at least a 3D LAAC device model onto at least a 3D cardiac model may include determining an optimal path for a placement of at least a 3D LAAC device model within at least a 3D cardiac model, generating a path model for an optimal path, and superimposing a path model onto at least a 3D cardiac model.

[0059] In one or more embodiments, a 3D cardiac model may include peripheral vasculature.

[0060] In one or more embodiments, generating at least a 3D cardiac model may include generating at least a 3D cardiac model using a point completion model.

[0061] In another aspect, a system for transesophageal echocardiogram-guided implantation of a left atrial appendage closure device includes at least a transesophageal echocardiogram system including at least an ultrasound sensor configured to be located within an esophagus of a patient and detect at least an ultrasound image as a function of cardiac tissue of the patient; at least a display; and at least a computing device configured to receive the at least an ultrasound image, input the at least an ultrasound image into a 3D reconstruction model, generate, using the 3D reconstruction model, at least a 3D cardiac model representative of a patient’s heart as a function of the at least an ultrasound image, receive at least a 3D watchman model representative of at least a left atrial appendage closure device, and display, using the at least a display, the at least an ultrasound image, the at least a 3D cardiac model, and the at least a 3D watchman model.

[0062] In one or more embodiments, a method of using a system to plan implantation of a left atrial appendage closure device may include providing feedback to an operator to ensure that at least an ultrasound sensor has been positioned and oriented to capture frames and views that are in sync with a Watchman IFU.

[0063] 8 Attorney Docket No. 1518-193PCT1 In one or more embodiments, a method may further include determining, using a placement and size determination engine, one or more of placement, including position and orientation, and size of a Watchman device.

[0064] In one or more embodiments, a method of using a system to guide implantation of a left atrial appendage closure device may include generating visual guidance and procedural feedback to assist an operator during an implantation procedure.

[0065] In one or more embodiments, a method of using a system post-implantation of a left atrial appendage closure device may include visualizing an actual implanted Watchman device using a post-implantation 3D reconstruction model.

[0066] In one or more embodiments, a method may further include displaying, using at least a display, a 3D constructed image of an actual implanted Watchman device and a 3D Watchman model overlaid with a recommended Watchman placement.

[0067] The details of one or more variations of the subject matter described herein are set forth in the accompanying drawings and the description below. Other features and advantages of the subject matter described herein will be apparent from the description and drawings, and from the claims.

[0068] DESCRIPTION OF DRAWINGS

[0069] For the purpose of illustrating the invention, the drawings show aspects of one or more embodiments of the invention. However, it should be understood that the present invention is not limited to the precise arrangements and instrumentalities shown in the drawings, wherein:

[0070] FIG. l is a diagram depicting an exemplary embodiment of an apparatus for determining a cardiac implant size;

[0071] FIG. 2 shows an exemplary embodiment of an intracardiac echocardiography (ICE) image;

[0072] FIG. 3 is a flow diagram of an exemplary embodiment of an ICE image example generation process;

[0073] FIG. 4 illustrates an exemplary embodiment of a three-dimensional (3D) voxel occupancy representation;

[0074] FIG. 5 is a schematic diagram of an exemplary transesophageal echocardiogram;

[0075] FIG. 6 presents 2D transesophageal echocardiogram (TEE) views at varying orientations;

[0076] FIG. 7 is a block diagram of an exemplary embodiment of a machine learning module;

[0077] 9 Attorney Docket No. 1518-193PCT1 FIG. 8 is a schematic diagram of an exemplary embodiment of a neural network;

[0078] FIG. 9 is a schematic diagram of an exemplary embodiment of a neural network node;

[0079] FIG. 10 is a flow diagram depicting an exemplary embodiment of a method of determining a cardiac implant size;

[0080] FIG. 11 illustrates a block diagram of an exemplary system for transesophageal echocardiogram-guided implantation of a left atrial appendage closure device;

[0081] FIG. 12 illustrates an exemplary display used during planning of implantation of LAAC device, according to some embodiments;

[0082] FIG. 13 illustrates an exemplary display used during implantation of LAAC device, according to some embodiments;

[0083] FIG. 14 illustrates a flow diagram showing an exemplary planning method;

[0084] FIG. 15 illustrates a flow diagram showing an exemplary implantation method;

[0085] FIG. 16 illustrates a flow diagram showing an exemplary post-implantation method;

[0086] FIG. 17 illustrates a flow diagram of an exemplary method for transesophageal echocardiogram-guided implantation of a left atrial appendage closure device; and

[0087] FIG. 18 illustrates a block diagram of a computing system that can be used to implement any one or more of the methodologies disclosed herein and any one or more portions thereof.

[0088] The drawings are not necessarily to scale and may be illustrated by phantom lines, diagrammatic representations, and fragmentary views. In certain instances, details that are not necessary for an understanding of the embodiments or that render other details difficult to perceive may have been omitted. Like reference symbols in the various drawings indicate like elements.

[0089] DETAILED DESCRIPTION

[0090] At a high level, an apparatus and method for determining a cardiac implant size is disclosed. An apparatus may include an ultrasonic imaging device and a computing device. Such a computing device may receive ultrasonic images of a subject’s heart, such as transesophageal echocardiograms (TEE) and / or intracardiac echo (ICE) images. A computing device may generate a 3D cardiac model by inputting ultrasonic images into a trained machine learning model. Such a machine learning model may be trained on a training dataset including ultrasonic images associated with 3D cardiac models generated based on historical computed tomography (CT) scan data. A 3D cardiac model generated by such a machine learning model may be used to

[0091] 10 Attorney Docket No. 1518-193PCT1 determine a cardiac measurement, which may be used to determine, as examples, a cardiac implant size, a cardiac implant placement, and / or a degree to which a subject is a suitable candidate for receiving a cardiac implant. In some embodiments, such an apparatus or method may represent an improvement over existing techniques, at least because CT scans, which take time and involve exposing the subject to radiation, are not required, and TEE may obtain clear ultrasonic images due to the proximity of a probe to the heart.

[0092] At a high level, aspects of the present disclosure are directed to systems and methods for transesophageal echocardiogram-guided implantation of a left atrial appendage closure device. The system includes at least a transesophageal echocardiogram (TEE) system including at least an ultrasound sensor, wherein the at least an ultrasound sensor is configured to be located within an esophagus of a patient and detect at least an ultrasound image as a function of cardiac tissue of the patient, at least a display and at least a computing device including at least a processor and a memory containing instructions configuring the at least a processor to receive the at least an ultrasound image, generate at least a 3D cardiac model representative of a heart of the patient as a function of the at least an ultrasound image, receive at least a 3D left atrial appendage closure (LAAC) device model representative of at least a LAAC device, generate a superimposed model by superimposing the at least a 3D LAAC device model onto the at least a 3D cardiac model and display, using the at least a display, the superimposed model.

[0093] Referring now to FIG. 1, an exemplary embodiment of an apparatus 100 for determining a cardiac implant size is illustrated. Apparatus 100 may include a computing device. Apparatus 100 may include a processor 104. Processor 104 may include, without limitation, any processor described in this disclosure. Processor 104 may be included in computing device. Computing device may include any computing device as described in this disclosure, including without limitation a microcontroller, microprocessor, digital signal processor (DSP) and / or system on a chip (SoC) as described in this disclosure. Computing device may include, be included in, and / or communicate with a mobile device such as a mobile telephone or smartphone. Computing device may include a single computing device operating independently, or may include two or more computing device operating in concert, in parallel, sequentially or the like; two or more computing devices may be included together in a single computing device or in two or more computing devices. Computing device may interface or communicate with one or more additional devices as described below in further detail via a network interface device. Network

[0094] 11 Attorney Docket No. 1518-193PCT1 interface device may be utilized for connecting computing device to one or more of a variety of networks, and one or more devices. Examples of a network interface device include, but are not limited to, a network interface card (c. ., a mobile network interface card, a LAN card), a modem, and any combination thereof. Examples of a network include, but are not limited to, a wide area network (e.g., the Internet, an enterprise network), a local area network (e.g., a network associated with an office, a building, a campus or other relatively small geographic space), a telephone network, a data network associated with a telephone / voice provider (e.g., a mobile communications provider data and / or voice network), a direct connection between two computing devices, and any combinations thereof. A network may employ a wired and / or a wireless mode of communication. In general, any network topology may be used. Information (e.g., data, software etc.) may be communicated to and / or from a computer and / or a computing device.

[0095] Still referring to FIG. 1, in some embodiments, apparatus 100 may include at least processor 104 and a memory 108 communicatively connected to the at least a processor 104, the memory 108 containing instructions 112 configuring the at least a processor 104 to perform one or more processes described herein. Computing device 116 may include processor 104 and / or memory 108. Computing device 116 may be configured to perform one or more processes described herein.

[0096] Still referring to FIG. 1, computing device 116 may include but is not limited to, for example, a computing device or cluster of computing devices in a first location and a second computing device or cluster of computing devices in a second location. Computing device 116 may include one or more computing devices dedicated to data storage, security, distribution of traffic for load balancing, and the like. Computing device 116 may distribute one or more computing tasks as described below across a plurality of computing devices of computing device, which may operate in parallel, in series, redundantly, or in any other manner used for distribution of tasks or memory between computing devices. Computing device 116 may be implemented, as a non-limiting example, using a “shared nothing” architecture.

[0097] Still referring to FIG. 1, computing device 116 may be designed and / or configured to perform any method, method step, or sequence of method steps in any embodiment described in this disclosure, in any order and with any degree of repetition. For instance, computing device 116 may be configured to perform a single step or sequence repeatedly until a desired or

[0098] 12 Attorney Docket No. 1518-193PCT1 commanded outcome is achieved; repetition of a step or a sequence of steps may be performed iteratively and / or recursively using outputs of previous repetitions as inputs to subsequent repetitions, aggregating inputs and / or outputs of repetitions to produce an aggregate result, reduction or decrement of one or more variables such as global variables, and / or division of a larger processing task into a set of iteratively addressed smaller processing tasks. Computing device 116 may perform any step or sequence of steps as described in this disclosure in parallel, such as simultaneously and / or substantially simultaneously performing a step two or more times using two or more parallel threads, processor cores, or the like; division of tasks between parallel threads and / or processes may be performed according to any protocol suitable for division of tasks between iterations. Persons skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various ways in which steps, sequences of steps, processing tasks, and / or data may be subdivided, shared, or otherwise dealt with using iteration, recursion, and / or parallel processing.

[0099] Still referring to FIG. 1, as used in this disclosure, “communicatively connected” means connected by way of a connection, attachment or linkage between two or more relata which allows for reception and / or transmittance of information therebetween. For example, and without limitation, this connection may be wired or wireless, direct or indirect, and between two or more components, circuits, devices, systems, and the like, which allows for reception and / or transmittance of data and / or signal(s) therebetween. Data and / or signals therebetween may include, without limitation, electrical, electromagnetic, magnetic, video, audio, radio and microwave data and / or signals, combinations thereof, and the like, among others. A communicative connection may be achieved, for example and without limitation, through wired or wireless electronic, digital or analog, communication, either directly or by way of one or more intervening devices or components. Further, communicative connection may include electrically coupling or connecting at least an output of one device, component, or circuit to at least an input of another device, component, or circuit. For example, and without limitation, via a bus or other facility for intercommunication between elements of a computing device. Communicative connecting may also include indirect connections via, for example without limitation, wireless connection, radio communication, low power wide area network, optical communication, magnetic, capacitive, or optical coupling, and the like. In some instances, the terminology “communicatively coupled” may be used in place of communicatively connected in this

[0100] 13 Attorney Docket No. 1518-193PCT1 disclosure.

[0101] Still referring to FIG. 1, processor 104 may include ultrasonic imaging device 120, and / or may be configured to collect plurality of ultrasonic images 124. Plurality of ultrasonic images 124 may include, without limitation, a two-dimensional image. As used herein, an “ultrasonic image” is an image generated as a function of a reflection of a sound wave off of a structure. As used herein, an “ultrasonic imaging device” is a device configured to collect ultrasonic images. Non-limiting examples of ultrasonic images and / or imaging techniques include intracardiac echo (ICE) images, transthoracic echocardiograms (TTE), transesophageal echocardiograms (TEE), and point of care ultrasound (POCUS). In some embodiments, plurality of ultrasonic images 124 may include a TEE image. In some embodiments, a TEE image may include a 2D TEE. In some embodiments, a TEE image may include a 3D TEE. In some embodiments, plurality of ultrasonic images 124 may include an ICE image. In some embodiments, a set of ultrasonic images of the patient’s organ may include an image selected from the list consisting of a transesophageal echocardiogram image, a transthoracic echocardiogram image, and a point-of- care ultrasound image. An ultrasonic image of plurality of ultrasonic images 124 may depict an organ and / or tissue of a subject. An ultrasonic image of plurality of ultrasonic images 124 may depict a heart, lung, spleen, liver, kidney, muscle, skeleton, intestine, stomach, vein, and / or artery. In some embodiments, an ultrasonic image may depict a heart, and / or a left atrium, left atrial appendage, left ventricle, right ventricle, and / or a right atrium of a heart.

[0102] Still referring to FIG. 1, in some embodiments, an ultrasonic image may include transesophageal echocardiogram (TEE) image. In some embodiments, a TEE image may include a view of an ostial diameter and / or length. In some embodiments, plurality of ultrasonic images 124 may include ultrasonic images captured at the same position, and different orientations with respect to a subject’s heart. In some embodiments, a TEE image may include a view of an ostial diameter and / or length from one or more angles on a mid-esophageal view. As examples, a TEE image may include a view of a subject’s heart and / or an ostial diameter and / or length from 1, 2, 3, 4, 5, 6, 7, 8, 9, or 10 different angles on a mid-esophageal view. In a non-limiting example, a TEE image may include a view of a subject’s heart and / or an ostial diameter and / or length from 4 different angles (0°, 45°, 90°, 135°) on a mid-esophageal view. In some embodiments, angles of images may describe a rotational position of an ultrasound transducer relative to a subject’s heart. In additional non-limiting examples, a TEE image may include a view of a subject’s heart

[0103] 14 Attorney Docket No. 1518-193PCT1 and / or an ostial diameter and / or length from 0° (which may provide a four chamber view), 45- 60° (which may provide a two chamber view and in some embodiments may be used to identify a thrombus), 90° (which may provide a long-axis view), 120-135° (which may provide a long- axis view of a left atrial appendage) and / or 135-160° (which may provide an aortic valve long axis view) on a mid-esophageal view. In some embodiments, an ostial diameter and / or length includes an ostial diameter and / or length of a left atrial appendage (LAA) of a heart and / or another location at which a cardiac implant may be placed. In some embodiments, TEE image may include a view of an ostial diameter and / or length from 4 different angles, such as, as a nonlimiting example, 0°, 45°, 90°, and 135°. TEE imaging is discussed further with respect to FIG. 5.

[0104] Still referring to FIG. 1, in some embodiments, an ultrasonic image may include an intracardiac echocardiography (ICE) image. As used herein, an “ICE image” is an ultrasound image obtained from within the heart’s chambers or blood vessels. In some cases, ICE images may be captured using a specialized catheter equipped with an ultrasound transducer that is inserted into the body and guided to the heart of subject. In an embodiment, an ultrasound image may provide a detailed and real-time visualization of cardiac anatomy. ICE images may also include internal structures, functions, and blood flow patterns of the heart of a subject. Plurality of ultrasonic images 124 may be related in terms of content, time of capture, sequence, or any other relevant parameters described herein. In a non-limiting example, each image of plurality of ultrasonic images 124 may represent a particular view, angle, or perspective of an object, subject, or scene, and may be in two-dimensional (2D) or 3D format. Images of plurality of ultrasonic images 124 may include, without limitation, any two-dimensional or three-dimensional images of any anatomy or anatomical structure, including without limitation images of any internal organ, tissue including without limitation muscular, connective tissue, epithelial tissue, and / or nervous tissue, bone, and / or any other element that may be imaged within a human and / or animal body.

[0105] Still referring to FIG. 1, in a non-limiting example, structures of a heart of a subject which may be imaged may include chambers (e.g., four chambers including left and right atria and left and right ventricles), valves (i.e., the structures that regulate blood flow between chambers and vessels, including mitral, tricuspid, aortic, and pulmonary valves), vessels (e.g., aorta, pulmonary arteries and veins, and coronary arteries), conduction system (i.e., a network of

[0106] 15 Attorney Docket No. 1518-193PCT1 specialized cells that control the heart’s electrical activity and rhythm), muscular and connective tissues (e.g., heart’s muscular walls, septa, any other connective tissues that provide structural integrity and enable contraction), LAA and other appendages, pathological features (e.g., any abnormalities, defects, and / or the like), and / or other components of a heart.

[0107] Still referring to FIG. 1, as used in this disclosure, a “subject” refers to an individual organism. In an embodiment, subject may include a human, such as a human in need of a cardiac implant. A cardiac implant may include a left atrial appendage occlusion (LAAO) device. A cardiac implant may include a watchman device. A watchman device may fit in the left atrium of a subject and form a barrier against blood clots in order to reduce the risk of stroke. Different patients may receive different size devices, and cardiac measurements made as described herein may be used to determine which size device to give to a particular subject. “Watchman device,” as used in this disclosure, is a left atrial appendage occlusion (LAAO) device configured for implantation within a patient’s left atrial appendage (LAA) to prevent thromboembolic events associated with atrial fibrillation. In one or more embodiments, watchman device may function as a mechanical barrier that seals LAA, thereby reducing the risk of blood clot formation and subsequent ischemic stroke. In one or more embodiments, watchman device may include a selfexpanding metallic frame covered by a permeable membrane or fabric, designed to conform to a patient’s unique atrial anatomy.

[0108] Still referring to FIG. 1, in an embodiment, each ultrasonic image of plurality of ultrasonic images 124 may include a particular view of subject’s heart’s chambers, valves, vessels, and / or the like. In a non-limiting example, plurality of ultrasonic images 124 may include multiple views e.g., different angles and perspectives of a subject’s heart. In another embodiment, plurality of ultrasonic images 124 may be arranged in a temporal sequence. In a non-limiting example, plurality of ultrasonic images 124 may include a series of images captured over time, allowing for an observation of dynamic cardiac functions such as beating, blood flow, and / or the like. In some cases, each ultrasonic image of plurality of ultrasonic images 124 may include a corresponding timestamp, wherein the timestamp may include an indicator showing a date and time of when the corresponding ultrasonic image was taken.

[0109] Still referring to FIG. 1, in some embodiments, plurality of ultrasonic images 124 may be received from an electronic health record database. For example, plurality of ultrasonic images 124 may be collected at a first point in time, stored in an electronic health record database, and

[0110] 16 Attorney Docket No. 1518-193PCT1 later used in further steps of a process described herein. In some embodiments, plurality of ultrasonic images 124 may be collected using an ultrasonic imaging device as part of a procedure for administering a cardiac implant to a subject.

[0111] Still referring to FIG. 1, in some embodiments, receiving plurality of ultrasonic images 124 may involve one or more image preprocessing steps. In some cases, processor 104 may be configured to calibrate one or more ultrasonic images of plurality of ultrasonic images 124 by correcting for distortions and / or ensuring accurate spatial representation of a heart of a subject. In a non-limiting example, processor 104 may select one or more reference objects within ultrasonic image which need calibration to correct spatial distortions. In some cases, processor 104 may be configured to place a phantom with pre-determined dimensions in an ultrasonic image and adjust ultrasonic image until the phantom’s dimensions are accurately represented. In another non-limiting example, an ultrasonic images’ brightness and / or contrast may be adjusted by processor 104 to ensure that echogenicity (reflectivity) of the tissues is accurately represented. One or more tissues with known echogenicity may be selected by processor 104 as reference tissues to adjust corresponding portions of the one or more ultrasonic images. In other cases, standardized correction curves may be applied in order to correct the echogenicity of ultrasonic images. As an ordinary person skilled in the art, upon reviewing the entirety of this disclosure, may be aware of various calibration techniques, such as, without limitation, temporal calibration, geometric calibration, among others that can be used by processor 104 to preprocess plurality of ultrasonic images 124.

[0112] Still referring to FIG. 1, receiving plurality of ultrasonic images 124 may include performing image segmentation on or more ultrasonic images of plurality of ultrasonic images 124. In some cases, image segmentation may include separating specific structures or regions of interest (ROI) from the background or other structures in a given ultrasonic image. In a nonlimiting example, processor 104 may be configured to use edge detection algorithms to outline the heart chambers, separating them from surrounding tissues. One or more filters may be applied to highlight the boundaries between different types of tissues during the segmentation. In another non-limiting examples, valves and vessels may also be segmented by applying thresholding techniques. Processor 104 may be configured to set an intensity threshold based on the known echogenicity of blood and vessel walls and select pixels or regions having intensity below or above the intensity threshold from the given ultrasonic image. In some cases, one or

[0113] 17 Attorney Docket No. 1518-193PCT1 more machine learning models may be used to perform image segmentations, for example, and without limitation, a U-net (i.e., a convolution neural network containing a contracting path as an encoder and an expansive path as a decoder, wherein the encoder and the decoder forms a U- shaped structure).

[0114] Still referring to FIG. 1, processor 104 may be configured to generate a 3D data structure representing a heart of a subject as a function of plurality of ultrasonic images 124. In a nonlimiting example, 3D data structure may include a 3D voxel occupancy representation (VOR). As used in this disclosure, a “3D voxel occupancy representation (VOR)” of anatomy is a 3D digital representation of a spatial structure of the anatomy, wherein the representation is composed of a plurality of discrete volumetric elements known as voxels. A “voxel,” for the purpose of this disclosure, is a 3D equivalent of a pixel in 2D imaging. While a pixel represents a point in a 2D image and may include properties such as color and / or brightness, a voxel may represent a volume in a 3D space and may include additional properties such density / occupancy as described below. In an embodiment, each voxel of plurality of voxels within 3D VOR may represent a specific portion of a heart. In some cases, voxel may be a smallest distinguishable box-shaped part (i.e., Ipx- Ipx- Ipx) of a three-dimensional image. In some cases, each voxel of plurality of voxels within VOR may be represented as a cube or rectangular prism (although other shapes may be used in specialized applications). Each voxel may include a size that determines a resolution of the 3D image or model. In an embodiment, smaller voxels may provide higher resolution; however, it may require more computational resources (e.g., RAM) for processor 104 to process.

[0115] Still referring to FIG. 1, each voxel of plurality of voxels within VOR may include one or more embedded values. As used herein, “embedded values” refers to specific numerical or categorical data associated with each voxel. In some cases, embedded values may represent various attributes or characteristics of the corresponding portion of structure that voxel represents. In a non-limiting example, embedded values may include density values, intensity values, texture information, or any other quantitative measures that provide insights into the underlying tissue. Such embedded values may be derived from set of ultrasonic images or other imaging modalities used to generate data structure. In some cases, embedded values may be utilized, by processor 104, to differentiate between different types of tissues, such as myocardial tissue, blood vessels, or chambers. Embedded values may also facilitate the visualization of

[0116] 18 Attorney Docket No. 1518-193PCT1 dynamic cardiac functions, for example, and without limitation, blood flow or heart beating by encoding temporal information such as timestamps within plurality of voxels.

[0117] Still referring to FIG. 1, in an embodiment, each voxel of plurality of voxels may include a presence indicator. As used in this disclosure, a “presence indicator” refers to a data element that indicates a presence or absence (i.e., occupancy) of tissue within that portion. In some cases, and without limitation, presence indicator may include an occupancy status as one of the embedded values described herein. Portion may include a specific location within 3D space where data structure is generated; for instance, and without limitation, a coordinate in 3D space represented in a tuple such as (x, y, z). In an embodiment, 3D VOR may provide a spatial framework that allows for the modeling and visualization of structure in 3D space. In some cases, 3D data structure may include a plurality of layers or slices (either horizontal [e.g., xy plane] or vertical [e.g., xz or yz plane depends on the view direction]), wherein each layer or slices of the plurality of layers or slices is corresponding to a different cross-sectional view of a structure of subject, and collectively forming a comprehensive 3D depiction of the structure. In a non-limiting example, 3D VOR having plurality of voxels with presence indicators may indicate whether each voxel in 3D space may be occupied by a part of a structure of subject. A binary value such as 0 or 1 may be configured as presence indicator to show either a pixel of 3D space is occupied (e.g., 1) or empty (e.g., 0). In should be noted that other values may be used as presence indicator such as a Boolean value e.g., TRUE or FALSE.

[0118] Still referring to FIG. 1, one or more embedded values, such as, without limitations, occupancy, or density, may be derived from plurality of ultrasonic images 124 described herein by processor 104. In a non-limiting example, determining occupancy status of each voxel of plurality of voxels may include converting set of ultrasonic images to a set of binary images and determining occupancy status of each voxel as a function of the structure of interest’s binary value. In some cases, occupancy status may include a value representing the likelihood of occupancy of the corresponding tissue. In another non-limiting example, density may be calculated, by processor 104, for each voxel as a function of the echogenicity of one or more pixels on a given ultrasonic image, wherein, the brightness of the given ultrasonic image may be analyzed since different tissues reflect ultrasound waves differently.

[0119] Still referring to FIG. 1, generating 3D data structure of a subject’s heart may include generating a 3D array. In some cases, processor 104 may divide 3D space into a grid of plurality

[0120] 19 Attorney Docket No. 1518-193PCT1 of voxels, each with specific x, y, and z coordinates as embedded values. Each element of 3D array may correspond to a voxel. In some cases, 3D array may allow for easy access and manipulation of plurality of voxels, enabling various analyses, visualizations, and transformations either described or not described herein. In a non-limiting example, embedded values may include a density of the tissue at a specific location of a patient’s body derived from one or more ultrasonic images of plurality of ultrasonic images 124.

[0121] Still referring to FIG. 1, 3D data structure of structure may include a 3D grid configured to map presence indicators and / or other embedded values described herein of plurality of voxels (e.g., tissue density, blood flow velocity, echogenicity or acoustic properties, and any other biophysical properties). As used in this disclosure, a “3D grid” refers to a 3D data structure that divides a given volume (e.g., volume of a structure) into a plurality of discrete units called cells (i.e., volume elements). In an embodiment, each cell within 3D grid may be associated with a distinct voxel. Mapping presence indicators or other embedded values may include assigning each presence indicator or embedded value to each point within 3D grid such as comers of each corresponding cell. Such values may be derived from plurality of ultrasonic images 124 as described above.

[0122] Still referring to FIG. 1, cells may be continuous, meaning that one or more cells may represent one or more continuous regions of space rather than discreate, separate units. In a nonlimiting example, instead of being uniform, mapped presence indicator and / or other embedded values may vary continuously across different cells or cell’s volume. In such embodiments, processor 104 may use interpolation to estimate other (unknown) embedded values within a range based on existing values such as known embedded values at specific points, thereby allowing for smooth transitions between cells. Exemplary interpolation methods may include, without limitation, linear interpolation, cubic interpolation, and / or the like. For example, and without limitation, if the corners of a cell have known values interpolation can be used to estimate the values at any point within the cell based on those corner values.

[0123] Still referring to FIG. 1, 3D data structure of a heart may include a 3D grid having a plurality of cells e.g., voxels, wherein each cell may contain a continuous range of values representing tissue density, blood flow velocity, or other properties (i.e., embedded values). Processor 104 may be configured to apply trilinear or tri cubic interpolation to estimate tissue density within each cell based on presence indicator or other known values at the cell’s

[0124] 20 Attorney Docket No. 1518-193PCT1 boundaries, since tissue densities change gradually; Such 3D grid may provide a smooth, continuous representation of heat’s internal structures, allowing for more nuanced analysis and visualization as described below. In a further embodiment, 3D grid with continuous cells may be additionally used in fluid dynamics simulations.

[0125] Still referring to FIG. 1, presence indicators and / or other embedded values may be mapped to a 3D grid as a function of array masking. In a non-limiting example, processor 104 may generate a mask e.g., a binary array that defines which voxels or cells are affected. Mask may be used to select or modify specific voxels or cells based on certain attributes; for instance, and without limitation, processor 104 may use a mask to isolate the LA within the heart focusing the analysis on that specific region. Such mask may include criteria defined by specific density thresholds that distinguish the LA’s tissue (i.e., voxels representing LA in 3D grid) from surrounding structures (i.e., neighboring voxels). In some cases, such mask may further include a binary mask, wherein each voxel in the 3D grid may be assigned a first presence indicator such as 1 if the voxel meets the criteria for the LA and a second presence indicator such as 0 if it does not. In some embodiments, mask may be directly applied to 3D grid, selecting, or modifying voxels or cells, thereby enabling processor 104 to highlight, exclude, or otherwise manipulate specific parts of a heart within 3D grid. Processor 104 may then perform an element-wise multiplication between 3D grid and the mask. Continuing from the previous non-limiting example, voxels corresponding to the LA (wherein the mask value is 1) may retain their original values, while other voxels (where the mask value is 0) may be set to 0 or other specific value (i.e., excluded or masked out).

[0126] Still referring to FIG. 1, in some embodiments, 3D grid may include one or more spatial features extracted from plurality of ultrasonic images 124. As used in this disclosure, “spatial features” are specific characteristics or attributes related to the spatial arrangement, shape, size, texture, or orientation of structures within a 3D space. In some cases, spatial features may include one or more embedded values described herein and their combinations thereof. In a nonlimiting example, spatial feature may be represented numerically as a vector, a metric or other mathematical constructs that capture specific spatial characteristics. In some cases, spatial features may also be visualized as contours, surfaces, or other geometric representations. In an embodiment, spatial features may be extracted using edge detection, texture analysis, or other image processing techniques (e.g., cleaning and enhancing images, image segmentation, and / or

[0127] 21 Attorney Docket No. 1518-193PCT1 the like). In another embodiment, one or more machine learning models, such as convolutional neural networks (CNNs) as described in further detail below, may be used to extract complex spatial features.

[0128] Still referring to FIG. 1, in a non-limiting example, one or more spatial features may include one or more shape features (i.e., characteristics related to the shape of specific structures), such as curvature, surface area, volume, and / or the like. In another non-limiting example, one or more spatial features may include one or more texture features (i.e., characteristics related to the texture or pattern within tissues, as seen in plurality of ultrasonic images 124), such as gray-level co-occurrence matrix (GLCM) features representing the texture of heart muscle tissue. In another non-limiting example, one or more spatial features may include one or more orientation features (i.e., characteristics related to the orientation or alignment of structures), such as the angle or alignment of the septum within the heart. In a further nonlimiting example, one or more spatial features may include one or more edge and boundary features (i.e., Characteristics related to the edges or boundaries between different structures), such as edge detection features highlighting the boundary between the myocardium and the cardiac chambers. As an ordinary person skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various spatial features extracted from plurality of ultrasonic images 124 consistent with this disclosure.

[0129] Still referring to FIG. 1, apparatus 100 may determine 3D cardiac model 128 as a function of plurality of ultrasonic images 124. As used herein, a “3D cardiac model” is a 3D representation of a subject’s heart. In some embodiments, apparatus 100 may determine 3D cardiac model 128 using 3D cardiac model generation machine learning model 132. 3D cardiac model generation machine learning model 132 may be trained to interpret ultrasonic images and / or generate 3D cardiac models 128 by learning relationships between ultrasonic images and corresponding computed tomography (CT) scan data. 3D cardiac model generation machine learning model 132 may be trained using a supervised learning algorithm. 3D cardiac model generation machine learning model 132 may include a neural network. 3D cardiac model generation machine learning model 132 may be trained on a training dataset 136 including example ultrasonic images 140, associated with example 3D cardiac models. Example ultrasonic images 140 may be generated based on historical CT scan data 144. For example, a 3D model generated based on CT scan data may be used to determine what ultrasonic images would

[0130] 22 Attorney Docket No. 1518-193PCT1 contain when taken from varying perspectives. Training dataset 136 may be obtained by, for example, associating historical ultrasonic images with historical CT scan based 3D cardiac models. Once 3D cardiac model generation machine learning model 132 is trained, it may be used to determine 3D cardiac model 128. Apparatus 100 may input plurality of ultrasonic images 124 into 3D cardiac model generation machine learning model 132, and apparatus 100 may receive 3D cardiac model from 3D cardiac model generation machine learning model 132.

[0131] Still referring to FIG. 1, in some embodiments, a training dataset may be generated by correlating an instance of computed tomography scan data with one or more historical ultrasonic images as a function of a medical record and a language model. For example, a language model may be used to interpret a medical record and / or determine whether an instance of computed tomography scan data should be associated with a historical ultrasonic image in a training dataset. For example, a language model may be used to interpret language of a medical record, and the output of the language model may be used to identify whether a medical event has taken place between when the historical ultrasonic image was taken and when the historical computed tomography scan data was recorded, such that they are not to be associated in a training dataset. In another example, a language model may be used to interpret language of a medical record, and the output of the language model may be used to identify whether historical ultrasonic image and historical computed tomography scan data were recorded in a sufficiently short time, such that they are associated in a training dataset. In some embodiments, a training dataset may be identified by generating a synthetic ultrasonic image as a function of an instance of computed tomography scan data.

[0132] Still referring to FIG, 1, in some embodiments, training dataset 136 may include 3D cardiac models including computed tomography (CT) based 3D model. As used in this disclosure, a “computed tomography (CT) based 3D model” is a 3D representation of a structure that is created using data from CT scans. In some embodiments, a computed tomography (CT) based 3D model may include a 3D representation of a structure and surrounding structures that is created using data from CT scans. Computed Tomography is a medical imaging technique that uses X-rays to capture cross-sectional images (slices) of the body. By taking a plurality of slices, a CT scan creates a detailed 3D representation of the internal structure. In an embodiment, CT- based 3D model may include 3D representations of the heart including chambers, valves, blood vessels, and surrounding tissues. CT-based 3D models may be generated using existing

[0133] 23 Attorney Docket No. 1518-193PCT1 techniques in the field as described above such as, without limitation, FAM, cardiac CT merging, among others.

[0134] Still referring to FIG. 1, processor 104 may be configured to receive at least an ultrasound localization datum. As used in this disclosure, an “ultrasound localization datum” is a unit of information that represents a position and / or angle of an ultrasound device. In some cases, ultrasound device may have incorporated one or more position sensors (e.g., accelerometers, gyroscope, inertial measurement units, magnetics location sensors, and the like) and / or be incorporated with an localization system. In some cases, one or more of ultrasound device and localization system communicates at least an ultrasound localization datum to processor 104. Processor 104 may be configured to generate, using 3D cardiac model generation machine learning model, the 3D cardiac model as a function of at least an ultrasound localization datum Still referring to FIG. 1, processor 104 may be configured to generate a 3D voxel occupancy representation (VOR) representing a cardiac shape as a function of plurality of ultrasonic images and 3D cardiac model generation machine learning model trained. In some versions, processor 104 may generate a mesh representing cardiac shape as a function of 3D voxel occupancy representation. Processor 104 may be configured to display, using display, a mesh to a user.

[0135] Still referring to FIG. 1, processor 104 may be configured to calculate a level of uncertainty at a plurality of locations on 3D cardiac model. In some cases, plurality of locations may include a high uncertainty region. Processor 104 may be configured to receive a subsequent plurality of ultrasonic images of cardiac anatomy corresponding to a high uncertainty region of 3D cardiac model. In some cases, subsequent plurality of ultrasonic images may be captured using ultrasonic imaging device (e.g., ICE, TEE, TTE, and / or POCUS), as a function of high uncertainty region. Processor 104 may be configured to generate a subsequent 3D cardiac model as a function of subsequent plurality of ultrasonic images, generating 3D model includes determining a level of uncertainty at least at one location of a plurality of locations of the 3D model based on the set of shape parameters. A location may refer to each voxel of plurality of voxels, cells, geometric marker, and all other identifying markers / data points of a model as described throughout this disclosure. A plurality of locations may refer to the surface of heart model, such as a set of pixels or a region on a model. “Uncertainty,” as used herein, refers to the lack of confidence or precision in a model's predictions. In some cases, the level of uncertainty

[0136] 24 Attorney Docket No. 1518-193PCT1 may be derived from variability within the distribution of shape parameters, image quality assessment, measurement errors and / or the like. In a non-limiting example, greater changes in heart geometry (indicated by the plurality of shape parameters) may correspond to a greater level of uncertainty at that location. This may be used to inform clinical decisions, for example, areas of high uncertainty may be avoided when planning a pathway for surgical intervention or additional imaging may be requested to reduce uncertainty in critical areas. Levels of uncertainty may refer to categories of uncertainty such as epistemic uncertainty, aleatoric uncertainty, model parameter uncertainty, pixel-wise uncertainty, boundary uncertainty, uncertainty in time series data, predictive uncertainty, systematic uncertainty, model output uncertainty, and the like. Epistemic uncertainty arises from a lack of knowledge or information. For example, limited training data for certain cardiac pathologies may contribute to higher epistemic uncertainty. Aleatoric uncertainty, also known as data uncertainty, results from inherent randomness or variability in the data. For example, variability in cardiac anatomy among different patients or imaging modalities introduces aleatoric uncertainty. Model Parameter Uncertainty is uncertainty associated with the model parameters, indicating how well the model has learned the underlying patterns in the training data. For example, variations in model parameters due to the stochastic nature of the optimization process contribute to parameter uncertainty. Pixel-wise Uncertainty is associated with individual pixels in the image. It provides a confidence measure for each pixel in the segmentation mask. For example, certain regions of the heart may be more challenging to segment accurately, leading to higher pixel-wise uncertainty. Boundary Uncertainty is related to the boundaries between different structures or regions in the image. For example, the precise delineation of the endocardium or epicardium may be uncertain in regions where the boundaries are not well-defined. Regarding uncertainty in Time Series Data, in tasks involving sequential data, such as cardiac imaging over time, uncertainty can be related to variations in the temporal dimension. For example, segmentation of dynamic structures like the beating heart involves handling uncertainty associated with different phases of the cardiac cycle. Predictive Uncertainty is uncertainty in the model's predictions for unseen data points. For example, when the model encounters a novel pathology or an atypical cardiac structure, predictive uncertainty measures its confidence in providing accurate segmentation. Systematic Uncertainty is uncertainty stemming from systematic errors or biases in the data collection process or the model architecture. For example, if the training data is biased towards a specific demographic, the model may exhibit

[0137] 25 Attorney Docket No. 1518-193PCT1 uncertainty when applied to a more diverse patient population. Model Output Uncertainty is uncertainty associated with the actual output of the model, indicating how confident the model is in its segmentation predictions. For example, the model may output a segmentation mask with a probability or confidence score for each pixel, reflecting the uncertainty associated with that pixel's classification. A level of uncertainty may include a degree, statistical measure, percentage, or variable whether linguistic or numerical, and the like identifying a range of uncertainty. For example, processor 104 may generate probability scores / confidence scores for locations of a model, indicating the model's confidence in its predictions. Calibration plots can be used to assess how well these confidence scores align with the true accuracy. Processor 104 may perform a threshold analysis to investigate how varying decision thresholds for classification or segmentation affects the trade-off between sensitivity and specificity in uncertain regions. Threshold analysis may include task-specific metrics for clinical relevance. For example, in cardiac image segmentation, critical regions like the myocardium may have stricter uncertainty thresholds compared to less critical regions. Processor 104 may implement Bayesian Neural Networks (BNNs) to perform posterior predictive checks to evaluate the agreement between the model's predictions and the observed data, such as data store, considering the uncertainty represented by the posterior distribution in Bayesian frameworks. In various embodiments, a level of uncertainty may be metrics determined by processor, such as Pixel-wise Uncertainty Metrics, Boundary Displacement Error (BDE), Uncertainty-Aware Loss Functions, Calibration Metrics, and the like. In some embodiments, level of uncertainty may be determined using Monte Carlo dropout. Monte Carlo dropout may include running a neural network multiple times using different dropout configurations. Each dropout configuration may include turning off turning off one or more nodes of a neural network. Monte Carlo dropout may be used to, for example, determine mean and variance parameters. In some embodiments, such a variance parameter may be used as level of uncertainty. In some embodiments, level of uncertainty may be determined using deep ensembles. A deep ensemble may include a plurality of machine learning models. An input may be applied to a plurality of machine learning model, and their outputs may be combined. For example, an average and / or variance of outputs of a plurality of models may be found. Level of uncertainty may be determined based on such variance. Shape identification model may be calibrated. Calibration may include fine-tuning or adjusting shape identification model predictions to align more closely with the actual probabilities. A well-

[0138] 26 Attorney Docket No. 1518-193PCT1 calibrated model is one where, for instance, if it predicts a 70% probability for a certain event, that event actually occurs about 70% of the time. Calibrating shape identification model may include receiving a set of validation data. As used in the current disclosure, a “set of validation data” is a set of data used to calibrate a machine learning model, which the machine learning model has not been trained on. A set of validation data is used to assess the model's performance and, in this case, to calibrate level of uncertainty. Processor 104 may sort each datapoint of the validation set into a plurality of hyperfine bins as a function of a continuous value. As used in the current disclosure, "hyperfine bins" is a grouping of the data points into a set of very fine or detailed bins. These bins may be organized based on the values of the continuous value. Each bin corresponds to a specific range or interval of continuous values. Processor 104 may additionally determine a bin-wise scaling factor for each of the plurality of hyperfine bins. As used in the current disclosure, "bin-wise scaling factors" refers to a factor or multiplier associated with each individual hyperfine bin. It may be used to adjust or scale the data within each bin. The scaling factor can be unique to each bin and is typically determined based on some specific criteria or algorithm. Calibration of level of uncertainty may include temperature scaling. Temperature scaling may include adjusting confidence scores or probabilities generated by a model to make them better reflect the true uncertainty or reliability of the model's predictions. This technique is often used to improve the calibration of deep neural networks, especially in cases where model confidence scores do not align well with actual probabilities. Temperature scaling introduces a hyperparameter known as the "temperature" (T). The temperature is a positive scalar value that is applied to the logits (raw scores) before they are passed through a SoftMax function. By adjusting the temperature, processor 104 may control the sharpness or spread of the probability distribution. A higher temperature makes the distribution more uniform, while a lower temperature makes the distribution sharper. In an embodiment, high temperature may smooth the distribution, reducing confidence in predictions. High temperature may increase level of uncertainty. Low temperature may sharpen the distribution, reducing level of uncertainty. Temperature parameter may be adjusted based on a validation dataset. A temperature that minimizes the difference between predicted probabilities and the true probabilities observed in a calibration dataset may be determined. In some embodiments, processor 104 may identify a high uncertainty location of 3D model. As used herein, a “high uncertainty location” is a location of a 3D model associated with a level of uncertainty that is above the average level of uncertainty of

[0139] 27 Attorney Docket No. 1518-193PCT1 the 3D model. A high uncertainty location may be determined using a method for determining uncertainty as described above. In a non-limiting example, a high uncertainty location may be determined using model output uncertainty. In some embodiments, a high uncertainty location may be identified on a map as described below. In some embodiments, a high uncertainty location may be determined using a method of determining a level of uncertainty used to generate a map as described below. In some embodiments, a high uncertainty location has higher level of uncertainty than 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, or 100% of locations of a 3D model. Processor 104 may be configured to generate a map regarding one or more levels of uncertainty. A “map,” as used herein, refers to a visualization. Map may include level(s) of uncertainty to be visualized on heart model. Map may include a color-coded heatmap, including other visual cues, symbols or indicators that alert a user to areas of 3D model that may require extra caution when used for planning or guidance during an ICE procedure. For example, after obtaining the segmentation results from heart model, map may be generated. Map may highlight the uncertainty or confidence level associated with each pixel in the segmentation. Assigning colors to different intensity levels in map allows for an intuitive visualization. Typically, warmer colors (e.g., red, or yellow) might represent high uncertainty, while cooler colors (e.g., blue, or green) could indicate low uncertainty. The color-coding can be adjusted based on specific thresholds or clinical requirements. In some embodiments, a map may include and / or identify one or more high uncertainty locations. For example, a high uncertainty location may be represented in a color such as red, orange, or yellow. In some embodiments, a mathematical formula may be used to generate colors based on uncertainty level. For example, a formula may convert high uncertainty values into hex values or RGB values associated with warmer colors and low uncertainty values into hex values or RGB values associated with cooler colors. Warm colors and cool colors need not be used to represent high and low uncertainty, respectively. For example, any first color and / or degree of opacity may represent high uncertainty, and any second color and / or degree of opacity may represent low transparency. For example, an opaque blue may be used to represent high uncertainty, and a moderately transparent yellow may be used to represent low uncertainty. In some embodiments, colors used may be those on a gradient from a first color to a second color. In some embodiments, only a single color may be used, such as when one of high or low uncertainty is represented by transparency. In some embodiments, more than two colors and / pr opacities may be used to represent uncertainty

[0140] 28 Attorney Docket No. 1518-193PCT1 values. For example, a first gradient may be used to represent uncertainty values in a first range, and a second gradient may be used to represent uncertainty values in a second range. In some embodiments, discrete colors are used for certain uncertainty ranges. In some embodiments, uncertainty values may be represented along a continuous range of color and / or opacity. In some embodiments, non-color features may be used to represent uncertainty values. In non-limiting examples, a particular pattern may be applied to an uncertain region, brightness or darkness may be used, and line width and dot size may be used. In some embodiments, visual representations used for uncertainty may be relative to one or more other points within that model. For example, the lowest uncertainty point may always be green, and the highest uncertainty point may always be red. This may allow a user to understand which points are the most or least uncertain, even when all points are relatively certain or uncertain. In some embodiments, a first visual representation of uncertainty may be used for absolute uncertainty, and a second visual representation of uncertainty may be used for relative uncertainty. For example, opacity may be used to represent relative uncertainty within a model, and color may be used to represent absolute uncertainty. In some embodiments, mathematical formula for converting the uncertainty values to colors may be recalculated and / or updated as a function of the upper and lower uncertainty values. For example, the mathematical formula may be updated such that the lowest uncertainty value represents one end of the color / opacity spectrum and the highest uncertainty value represents the other end of the color / opacity spectrum. Generating map may include methods such as Class Activation Mapping (CAM). Class Activation Mapping is a technique that originated for image classification tasks and has been extended to provide visual insights into the regions of an image that are most important for a particular class. CAM allows the visualization of the spatial attention of a convolutional neural network (CNN) by generating heat maps that highlight discriminative regions. CAM may be applied to the last convolutional layer of a CNN. The features extracted by this layer capture high-level semantic information, making it suitable for visualizing the importance of different regions in an image. CAM is typically applied to the last convolutional layer of a CNN. The features extracted by this layer capture high-level semantic information, making it suitable for visualizing the importance of different regions in an image. The output of the global average pooling is then fed into a fully connected layer with a softmax activation function. This converts the features into class scores, indicating the likelihood of the image belonging to different classes. The CAM algorithm computes a weighted sum of the

[0141] 29 Attorney Docket No. 1518-193PCT1 original feature maps based on the weights of the fully connected layer. These weights are determined during the training process and represent the importance of each feature map for a specific class. The weighted sum is applied to the original feature maps, producing a single heat map. This heat map highlights the regions of the input image that contributed most to the prediction for the target class. The generated heat map can be overlaid on the input image, visually indicating which regions are most relevant for the predicted class. Typically, warmer colors (e.g., red, or yellow) represent higher activation or importance. Generating map may include Grad-CAM (Gradient-weighted Class Activation Mapping). Grad-CAM is an extension of Class Activation Mapping (CAM) that enhances the localization capabilities by incorporating gradient information from the final convolutional layer of a neural network. Grad-CAM helps to generate heat maps that highlight discriminative regions in an image, providing more finegrained insights into where a convolutional neural network (CNN) is focusing its attention when making predictions. In traditional CAM, the last convolutional layer's feature maps are linearly combined to obtain a weighted sum, and the resulting weights are used to create a heat map that highlights relevant regions for a specific class. Grad-CAM improves upon CAM by introducing gradient information. It computes the gradients of the predicted class score with respect to the feature maps of the last convolutional layer. Grad-CAM retains the global average pooling (GAP) operation applied after the last convolutional layer, as it is an integral part of CAM. The GAP operation condenses the spatial information into a single value per feature map. The gradients obtained in the previous step are used to calculate the importance of each feature map. These gradients represent the importance of each feature map in contributing to the final prediction. A weighted sum is computed using these gradients, and this is combined with the original feature maps. The computed sum goes through a ReLU activation function, discarding any negative values. This step emphasizes positive contributions and suppresses negative ones. The ReLU -activated weighted sum is linearly combined with the original feature maps to produce a weighted combination. This combination retains spatial information and helps create a more accurate heat map. The resulting heat map is often normalized to enhance visualization, ensuring that the values are within a specific range (e.g., between 0 and 1). The final heat map generated by Grad-CAM is then overlaid on the input image, highlighting the regions of interest for the predicted class. The intensity of the heat map indicates the importance of different regions. Grad-CAM enhances the interpretability and explainability of deep learning models,

[0142] 30 Attorney Docket No. 1518-193PCT1 allowing practitioners and researchers to understand which parts of an image are crucial for a particular prediction. This is particularly valuable in applications such as medical imaging or any domain where understanding the decision-making process is critical. Generating map may include utilizing a “SmoothGrad technique,” a technique designed to improve the interpretability of neural network predictions by reducing the noise in the attribution maps or heat maps generated by visualizing gradients. It is particularly useful for understanding the decision-making process of deep learning models, especially in scenarios where the explanations need to be robust and less sensitive to input perturbations. The primary goal of SmoothGrad is to enhance the visual quality of attribution maps generated by visualizing gradients. Attribution maps highlight the regions in the input that contribute most to a model's prediction. SmoothGrad aims to reduce the impact of noise in these maps, providing more stable and interpretable visualizations. The key idea behind SmoothGrad is to introduce perturbations to the input data. Instead of attributing the prediction solely to the gradients calculated with respect to the original input, the gradients are averaged over multiple perturbed versions of the input. By averaging the gradients over multiple perturbed samples, SmoothGrad helps reduce the impact of noise or irrelevant features in the attribution maps. This is particularly beneficial when dealing with complex or noisy datasets. Perturbation techniques include adding Gaussian noise, random rotations, or random translations to the input data. These perturbations create variations in the input while preserving the essential features, leading to more stable and reliable attribution maps. For each perturbed input, gradients are calculated with respect to the model's output. These gradients are then averaged over all perturbed samples. This process smoothens the attribution map by reducing the influence of random noise. The averaged gradients may undergo normalization or scaling to ensure that the values are interpretable and within a specific range. This step can enhance the consistency and comparability of the generated attribution maps. The final step involves generating a heat map using the smoothed gradients. The heat map represents the attribution of different regions in the input to the model's prediction, providing a clearer and more stable visualization. Generating map may include implementing one or more Gaussian Processes. A Gaussian Process is a collection of random variables, any finite subset of which has a joint Gaussian distribution. In simpler terms, it is a distribution over functions rather than a distribution over finite-dimensional vectors. Gaussian Processes (GPs) can be applied to generate heat maps in various ways, particularly in the context of regression tasks where one would want

[0143] 31 Attorney Docket No. 1518-193PCT1 to predict continuous values across a spatial domain. Given a set of observed data points, the GP can predict the values at unobserved locations in the spatial domain. Importantly, it also provides uncertainty estimates associated with these predictions. This uncertainty can be visualized as a heat map. The predicted values from the GP represent the main heat map, indicating the expected values across the spatial domain. The uncertainty associated with each prediction can be visualized as an uncertainty heat map. This uncertainty heat map provides insights into regions where the model is less confident about its predictions. Overlay of the main heat map and the uncertainty heat map on the original spatial data may create a composite visualization. Warmer colors in the main heat map might represent higher predicted values, while the uncertainty heat map's intensity could indicate regions where the model's predictions are less certain. Uncertainty may be depicted in ways other than a color coded heatmap. For example, a 2-dimensional cross section of a 3D model may be taken, and a third dimension indicating certainty or uncertainty may be added. In another example, transparency may indicate certainty or uncertainty. In another example, sharpness may indicate certainty or uncertainty. In another example, size of dots or thickness of lines making up a 3D model may indicate certainty or uncertainty. In some embodiments, locations with particular certainty features may be identified. For example, high uncertainty location may be included and / or identified. Processor 104 may be configured to overlay map onto heart model. In some embodiments, the overlay may be placed on 3D model and go through a refinement process as described above. Overlaying map on a model may include implementing spatial alignment methods, texture mapping techniques wherein the color information from the heat map is mapped onto the vertices or faces of the 3D model, shader programs that define how the heat map values influence the final appearance of the 3D model, visualization software or programming libraries that support 3D rendering and overlay capabilities, interactivity visualization, quality control methods, and the like. For example, texture mapping may include UV Mapping wherein each point on the surface of a 3D model is associated with a set of texture coordinates often denoted as U and V. These coordinates are analogous to the X and Y coordinates on a 2D image. UV mapping establishes the correspondence between points on the 3D model and pixels on the 2D texture. In another example, interactive visualization may create visual representations of data that users can interact with and manipulate. This approach allows users to explore and analyze data dynamically, gaining insights through direct engagement with the visual representation. For example, mouse

[0144] 32 Attorney Docket No. 1518-193PCT1 interactivity may allow users to interact with visual elements using mouse actions, such as hovering over data points for additional information, clicking to drill down into details, or dragging to pan and zoom. Filtering and Selection capabilities may allow a user to filter data based on specific criteria or select subsets of data for closer examination. This is particularly useful when dealing with large datasets. Spatial Exploration may allow users to zoom in to explore details or pan to navigate across the space. In some cases, an ICE frame taken during an ICE procedure or synthesized for machine learning training purposes may be also overlaid at a corresponding location or heart model. Overlaying the ICE frame may include registering the ICE frame to the generated 3D model using the image processing model. For example, the processing system may include at least a processor 104 and a memory communicatively connected to at least a processor 104, wherein the memory contains instructions configuring at least a processor 104 to receive a heart model related to a patient’s heart, identify a region of interest within the heart model, wherein identifying the region of interest includes locating at least a point of view on the heart model and determining a view angle corresponding to the at least a view origin, wherein the at least a point of view and the corresponding view angle define at least one field of view that include at least a portion of the heart model. At least a processor 104 may be further configured to generate at least a medical image as a function of the region of interest using an image generator, wherein the at least a medical image captures an anatomical structure of the at least a portion of the heart model. In some embodiments, processor 104 may receive more than one set of images of cardiac anatomy. For example, processor 104 may receive a first set of images of cardiac anatomy, and such first image may be used to perform one or more functions described herein. As examples, a 3D model, a level of uncertainty, and / or a map may be generated based on such first set of images. A second set of images may be captured as a function of such a 3D model, level of uncertainty, and / or map. In some embodiments, a second set of images may be captured of one or more high uncertainty locations. Image sets beyond the first may be captured and / or received as described above with respect to set of images. A second set of images may be captured as a function of a high uncertainty location. For example, a first set of images may be used to determine a first 3D model, and a map including a high uncertainty location may be overlaid on the first 3D model. Such first 3D model and map may be displayed to a user, such as through a display device. A user, such as a medical professional, may use a cardiac image capture device to capture a second set of images based on

[0145] 33 Attorney Docket No. 1518-193PCT1 the 3D model and map. For example, a user may use a cardiac image capture device to capture a second set of images of a high uncertainty location. Image may set beyond the first may be used to create a 3D model and / or a map. In some embodiments, a second set of images may be captured, and the second set of images may be used to generate a 3D model and / or a map as described herein. In some embodiments, a second set of images may be combined with an earlier captured first set of images to generate a 3D model and / or a map. For example, a first set of images may provide data as to a majority of a cardiac anatomy, but a 3D model based on a first set of images may include one or more high uncertainty locations; a second set of images providing additional data on one or more high uncertainty locations may be captured by a user using a cardiac image capture device, and the first set of images and second set of images may be combined to provide more complete cardiac anatomy data. A second 3D model and / or a second map may be generated based on the first set of images and the second set of images. In some embodiments, a second 3D model may be generated using a filtered set of first and second images, as disclosed in this disclosure. In some embodiments, processor 104 may continuously update second 3D model as further images are collected. Updating second 3D model may include any of the methods for generating a 3D model as disclosed in this disclosure. Processor 104 may be configured to update a 3D after every set time period. Set time period may include 1 second, 5 seconds, 10 seconds, 1 minute, and the like. Set time period may range from 10 ms to 5 minutes. Processor 104 may be configured to update a 3D model after a certain number of additional images have been collected. For example processor 104 may update 3D model after 20 additional images have been collected. For example processor 104 may update 3D model after 10 additional images have been collected. For example processor 104 may update 3D model after 50 additional images have been collected. For example processor 104 may update 3D model after 100 or more additional images have been collected. In some embodiments, a second set of images may be captured based on first set of images. For example, if levels of uncertainty of a 3D model are sufficiently high, then processor 104 may configure a display device to display a notification to a user indicating that a second set of images is to be captured. In another example, if levels of uncertainty of a 3D model are sufficiently high, then a user may capture a second set of images. In some embodiments, processor 104 may automatically trigger capture of a second set of images and / or repositioning of a cardiac image capture device. In some embodiments, this may occur based on a level of uncertainty of a 3D model. For example,

[0146] 34 Attorney Docket No. 1518-193PCT1 processor 104 may automatically reposition a cardiac image capture device into a location and orientation for capturing an image of a part of cardiac anatomy associated with a high uncertainty location of 3D model. This may occur, for example, if a level of uncertainty exceeds a threshold. Non-limiting metrics which may be used to determine information displayed to a user and / or automatically capture a second set of images include a maximum uncertainty, an average uncertainty, a median uncertainty, a level of uncertainty at a certain uncertainty percentile, and the like. Such metrics may be compared to thresholds to determine information displayed to a user and / or automatically capture a second set of images.

[0147] Still referring to FIG. 1, processor 104 may be configured receive a 3D cardiac implant model representing a cardiac implant. As used in this disclosure, a “3D cardiac implant model” is a three-dimensional representation (i.e., image) of a cardiac implant. In some cases, 3D cardiac implant model may be represented aspe model, a mesh model, a point cloud, a VOR, or the like. In some cases, a 3D cardiac model may be generated according to any machine learning process described in this disclosure. In some cases, 3D cardiac implant model may be downloaded from a repository. Processor 104 may be configured to display, using display, 3D cardiac implant model with 3d cardiac model. In some cases, processor 104 may be configured to position 3D cardiac implant model relative 3D cardiac model. Processor 104 may be configured to receive a query image. For the purposes of disclosure, a “query image” is an image used as a query to match another image and / or to selectively retrieve information for use in further method steps as disclosed below; each query image has an associated region of interest (ROI) that is to be determined or estimated in 3D space, as described below. Query image may include a medical image. For the purposes of this disclosure, a “medical image” is a two- dimensional visual representation containing information pertaining to an interior of a body and functions of organs / tissues therein that may aid clinical analysis and medical intervention. Query image may include, without limitation, X-ray image, echocardiogram (ECG), magnetic resonance imaging (MRI) scan, computed tomography (CT) scan, ultrasound image including intracardiac echocardiogram (ICE) frame, transthoracic echocardiogram (TTE) frame, magnetic resonance imaging (MRI) scan, and / or transesophageal echocardiogram (TEE) frame, optical image, digital photograph, and / or the like. For the purposes of this disclosure, computed tomography (CT) is a medical imaging technique that uses X-rays to capture cross-sectional images (slices) of a patient’s body; by taking a plurality of slices, a CT scan creates a detailed

[0148] 35 Attorney Docket No. 1518-193PCT1 three-dimensional (3D) representation of internal structures. For the purposes of this disclosure, an “ICE frame” is a 2D ultrasound image that represents anatomy (i.e., walls, chambers, blood vessels, etc.) of at least part of a heart, as described above. For the purposes of this disclosure, a “transthoracic echocardiogram (TTE) frame” is a two-dimensional (2D) ultrasound image collected by placing a probe or ultrasound transducer on patient’s chest or abdomen to collect various views of heart. For the purposes of this disclosure, a “transesophageal echocardiogram (TEE) frame” is a 2D ultrasound image collected by passing a specialized probe containing an ultrasound transducer at its tip into patient’s esophagus; it is an alternative way of performing echocardiography. For the purposes of this disclosure, “echocardiography” is an imaging technique that uses ultrasound to examine heart, the resulting visual image of which is an echocardiogram. Anatomical structures may include, without limitation, chambers (e.g., four chambers including left and right atria and left and right ventricles), valves (i.e., the structures that regulate blood flow between chambers and vessels, including mitral, tricuspid, aortic, and pulmonary valves), vessels (e.g., aorta, pulmonary arteries and veins, and coronary arteries), conduction system (i.e., a network of specialized cells that controls the heart’s electrical activity and rhythm), muscular and connective tissues (e.g., heart’s muscular walls, septa, any other connective tissues that provide structural integrity and enable contraction), left atrial appendage and other appendages, pathological features (e.g., any abnormalities, defects, and / or the like), among others. ICE frame may be either collected and recorded by a medical professional using an image capture device, such as an ICE catheter, or synthesized from a 3D model using a synthetic ICE generator, as described below. In one or more embodiments, query image may be saved to and / or retrieved later from a patient profile and / or a database. For the purposes of this disclosure, a “model” or “3D model” refers to a digital representation of a three-dimensional object, capturing its internal structures and geometry. In one or more embodiments, 3D model may be a digital representation (i.e., a 3D heart model) of a patient’s heart, capturing its anatomy, geometry, and potentially functional properties. The apparatus and methods described in this disclosure may be agnostic to how 3D model is generated. As nonlimiting examples, 3D heart model may be generated from electro-anatomical mapping, pre-operative computed tomography (CT), MRI scans, or synthetically reconstructed using echocardiography frames such as, without limitation, ICE frames and transthoracic echocardiogram (TTE) frames. For the purposes of this disclosure, a patient is a human or any individual organism, on whom or on

[0149] 36 Attorney Docket No. 1518-193PCT1 which a procedure, study, or otherwise experiment, such as without limitation, atrial fibrillation ablation, may be conducted. In a nonlimiting example, processor 104 may receive model from a human patient with atrial fibrillation who is undergoing a procedure, an individual undergoing cardiac screening, a participant in a clinical trial, an individual with congenital heart disease, a heart transplant candidate, an individual receiving follow-up care after cardiac surgery, a healthy volunteer, an individual with heart failure, or the like. Additionally or alternatively, patient may include an animal model (i.e., an animal used to model atrial fibrillation such as a laboratory rat). In one or more embodiments, model may be received from a statistical shape model. For the purposes of this disclosure, a “statistical shape model (SSM)” is a data structure representing, including, and / or utilizing a mathematical model that captures principal modes of variation in shape across a population of similar three-dimensional structures, such as cardiac anatomies. SSM may capture a plurality of models associated with a plurality of patients. In one or more embodiments, SSM may be used to capture the variability in anatomical structures among different patients; for instance, SSM of a human heart may be constructed from a plurality of heart images collected from a plurality of individuals. In one or more embodiments, when model represents a heart, the model generated from SSM may capture an “average” heart shape and main ways in which heart shapes may vary among plurality of individuals. In one or more embodiments, when model and / or SSM represents a heart, segmentation of the heart may include a plurality of pixel values, e.g., 0-255, each representing a presence of heart tissue at that location. In a nonlimiting example, computer vision module may be configured to generate a mesh representation of a patient’s heart based on plurality of CT scan segmentations or other image segmentations, wherein the mesh representation may include a 3D VOR, as described above, using Pix2Vox. Additionally or alternatively, exemplary computer vision tasks may include, without limitation, object recognition, feature detection, edge / corner detection, and the like. Nonlimiting examples of feature detection may include scale invariant feature transform (SIFT), canny edge detection, Shi Tomasi comer detection, and / or the like. In one or more embodiments, generating mesh representation of patient’s heart may include employing, by computer vision module, one or more transformations to orient one or more images with respect to a 3D coordinate system; exemplary transformations include without limitation homography transforms and affine transforms. Computer vision module may implement one or more 3D modeling algorithms, such as without limitation, marching cubes, contour detection and

[0150] 37 Attorney Docket No. 1518-193PCT1 segmentation, active contour models, and / or the like to generate a coherent 3D representation based on mesh representation of an object, e.g., model. In one or more embodiments, generic 3D modeling techniques may be applied by computer vision module to generate model. In one or more embodiments, generic 3D modeling techniques may include surface modeling, solid modeling, parametric modeling, among others. A person of ordinary skill in the art, upon reviewing the entirety of this disclosure, will be aware of various image processing, computer vision, and modeling tasks that may be performed by processor 104 to generate model from a set of images such as heart images. In one or more embodiments, voxel may be a smallest distinguishable box-shaped part (i.e., Ipx x Ipx x Ipx) of 3D representation of heart. In one or more embodiments, each voxel within a plurality of voxels in 3D VOR may be represented as a cube or rectangular prism (although other shapes may be used in specialized applications). Each voxel may include a size that determines the resolution of a 3D model. In one or more embodiments, smaller voxels may provide higher resolution; however, it may require more computational resources (e.g., RAM) for processor to process. In one or more embodiments, each voxel may include one or more embedded values (i.e., specific numerical or categorical data associated with each voxel). In one or more embodiments, embedded values may represent various attributes or characteristics of the corresponding portion of heart that voxel represents. In a nonlimiting example, embedded values may include density values, intensity values, texture information, or any other quantitative measures that provide insights into the underlying content (e.g., tissue). In another nonlimiting example, each voxel may include a presence indicator, i.e., a data element that indicates a presence or absence (i.e., occupancy) of content within a portion of an object (e.g., heart), as described in U.S. Pat. App. Ser. No. 18 / 376,688. Such embedded values may be derived from corresponding labels of example shape. Model may include a plurality of regions of interest (ROIs) ; in one or more embodiments, each ROI within the plurality of ROIs may correspond to one query image and may be specified when the query image is matched to a corresponding synthetic image within a synthetic image repository, as described below. For the purposes of this disclosure, a “region of interest (ROI)” is a specific and pre-defined spatial subset of an image or a 3D model. In one or more embodiments, ROI may include a volume that has been designated for closer analysis or further processing as described in detail below due to its potential significance or relevance in synthesizing images. In one or more embodiments, identifying ROI within model may include isolating ROI from surrounding structure or structures

[0151] 38 Attorney Docket No. 1518-193PCT1 that may be less relevant. In one or more embodiments, ROI may be manually selected by user. In one or more embodiments, one or more graphical tools and / or imaging software may be used to outline a particular area within model or an image captured from model. In one or more embodiments, processor 104 may be configured to automatically detect and define ROI. In one or more embodiments, a computer vision module configured to perform one or more computer vision tasks such as, without limitation, thresholding, edge detection, or machine learning process may be used to recognize ROI with specific features or anomalies. In one or more embodiments, ROI may also include temporal ROI. In one or more embodiments, ROI may be not only spatial but also temporal. In one or more embodiments, a specific timeframe within a sequence may be designated as a ROI. In a nonlimiting example, temporal ROI may focus on a specific time segment or interval within a dynamic dataset, e.g., model, with an animation that simulates a cardiac cycle. In one or more embodiments, temporal ROI may change over time. For example, and without limitation, temporal ROI may include a time-series images capturing patient’s heart activity, or a sequence showcasing blood flow within the cardiac structure. In a nonlimiting example, ROI may include temporal ROI set to capture a specific phase of cardiac cycle such as systole or diastole. In one or more embodiments, ROI may include a hierarchical ROI. In a nonlimiting example, processor 104 may identify one or more smaller sub-ROIs within a larger ROI, each with its significance or weight. In or more embodiments, ROI may include a at least a field of view. Each field of view may include at least a portion of model and / or may further include at least a point of view and at least a view angle. For the purposes of this disclosure, a “point of view” is a specific spatial location or origin form which an image or scene is observed or captured. In a nonlimiting example, point of view may be configured to mimic the location of an image capture device such as ICE catheter, within or near patient’s heart. In one or more embodiments, at least a point of view may be imagined as the location of a virtual image capture device. In one or more embodiments, at least a point of view may determine from where within model or its vicinity “pseudo” ultrasound waves are emitted and / or received. Given that ICE is a type of endoluminal ultrasound, in one or more embodiments, at least a point of view may be intracardiac and located inside heart chambers. Exemplary point of views may include, without limitation, ventricular point of view, atrial point of view, near-valvular point of view, and / or the like. In a nonlimiting example, ROI may be identified and at least a point of view may be located on the left ventricle’s wall, targeting its thickness and motion to assess potential

[0152] 39 Attorney Docket No. 1518-193PCT1 cardiomyopathy. For the purposes of this disclosure, a “view angle” is an angular orientation or direction (i.e., defined by one or more 9 and (p angles within spherical coordinates) associated with and projected from at least a point of view. In one or more embodiments, view angle may determine the segment of a scene or image that is visible or captured. In a nonlimiting example, view angle may reflect the orientation of an imaging plane relative to the structure of interest within identified ROI. In one or more embodiments, view angle corresponding to at least a point of view may define the tilt of the imaging plane, determining which structures come into field of view. In one or more embodiments, field of view may indicate an area of a scene that may be captured by image capture device within defined bounds (e.g., spatial boundary of ROI) inside model. Exemplary view angle may include apical view (visualize patient’s heart from its apex), parasternal view (oriented laterally from the mid-sternal line), subcostal view (with angle inferiorly positioned). In one or more embodiments, view angle may correspond to the angle of the sector of a resultant medical image, such as an ICE image as described in detail below (which resembles a sector or-pie slice shape), wherein an ICE catheter tip may act as the sector’s apex (i.e., point of view ) that delineates an ultrasound wave’s spread and hence, the width of captured anatomy. In a nonlimiting example, a narrower view angle may be chosen to focus on a specific region of patient’s heart e.g., a valve. Conversely, a broader view angle may capture a more extensive heart region, offering a comprehensive overview of model. In one or more embodiments, processor 104 may be further configured to query a synthetic image repository for at least a matching synthetic image based on extracted neural network encodings of query image. Synthetic image repository includes plurality of synthetic images (e.g., a plurality of synthetic ICE frames, as described in detail below), and neural network encodings are extracted from each synthetic image by following the same or similar procedures as described above for query image. Synthetic image repository may be implemented in any manner suitable for implementation of database, as described in this disclosure. Each synthetic image has a one-to-one correspondence with ROI, field of view, point of view, and / or view angle within 3D model. In one or more embodiments, each synthetic image may be stored in database alongside its corresponding neural network encodings, ROI, field of view, point of view, and / or view angle. Querying synthetic image repository involves comparing extracted neural network encodings of query image with extracted neural network encodings of each synthetic image within the plurality of synthetic images. For the purposes of this disclosure, a “matching” synthetic image is a synthetic image

[0153] 40 Attorney Docket No. 1518-193PCT1 with the same neural network encodings (i.e., embeddings or vectors, as described above), the same overall geometric features, and the same pattern of organization between elements therein as query image. In one or more embodiments, for certain query image, there may only be “near matches” instead of exact matches (e.g., based on matching vectors between extracted neural network encodings) within synthetic image repository. In one or more embodiments, when only one near match is detected, processor 104 may be configured to sample around the near match (e.g., within a certain threshold distance and / or angle) to identify an exact match. In one or more embodiments, when two or more near matches are detected, processor 104 may be configured to interpolate between the two or more near matches to identify a 2D projection that is an exact match. Details regarding how 2D projections may be generated are described above in this disclosure. Processor 104 may be configured to display an estimated ROI of query image within 3D model. For the purposes of this disclosure, an estimated ROI is an approximate fraction within 3D model that significantly overlaps with the actual ROI for query image with a reasonable level of certainty; in other words, an estimated ROI may deviate slightly from actual ROI, but the deviation is minor enough ensure the overall precision of apparatus 100 during its operation, e.g., a medical procedure. In one or more embodiments, estimated ROI may be a function of ROI associated with matching synthetic image from query. In one or more embodiments, displaying estimated ROI may include overlaying a 2D cross section within at least a portion of 3D model, as described below. In one or more embodiments, estimated ROI and / or elements related thereto may be displayed on one or more display devices, as described below. Apparatus 100 may further include or be coupled to at least a display device. For the purposes of this disclosure, a “display device” is an electronic device that visually presents information to user. In one or more embodiments, display device may include an output interface that translates data such as, without limitation, synthetic image, from processor or other computing devices into a visual form that can be easily understood by user. In one or more embodiments, synthetic image and / or other data described herein such as, without limitation, model, patient profde, and / or the like may also be displayed through display device using a user interface. User interface may include a graphical user interface (GUI), wherein the GUI may include a window in which query image, plurality of synthetic image, among other data described herein, may be displayed. In one or more embodiments, user interface may include one or more graphical locator and / or cursor facilities allowing user to interact with query image,

[0154] 41 Attorney Docket No. 1518-193PCT1 synthetic image, and / or any other data, or even process described herein; for instance, and without limitation, by using a touchscreen, touchpad, mouse, keyboard, and / or other manual data entry device, user may enter user input containing selecting specific regions, adding comments, adjusting parameter, and / or the like. In a nonlimiting example, user interface may include one or more menus and / or panels permitting selection of measurements, models, visualization of data / model to be displayed and / or used, elements of data, functions, or other aspects of data / model to be edited, added, and / or manipulated, options for importation of and / or linking to application programmer interfaces (APIs), exterior services, data source, machine-learning models, and / or algorithms, or the like. A person of ordinary skill in the art, upon reviewing the entirety of this disclosure, will be aware of various ways in which a visual interface and / or elements thereof may be implemented and / or used as described in this disclosure.

[0155] Still referring to FIG. 1, processor 104 may be configured to generate a set of shape parameters based on plurality of ultrasonic images 124. As used in this disclosure, a “set of shape parameters” refers to a collection of numerical values or descriptors that quantitatively represent the geometric or morphological characteristics of a heart. In some embodiments, a set of shape parameters may represent a shape of a heart. In a non-limiting example, set of shape parameters may include information and / or metadata calculated, determined, and / or extracted from set of ultrasonic images, such as, dimensions, angles, curvatures, surface areas, texture, symmetry, and / or the like. In other embodiments, processor 104 may be configured to parameterize features (e g., edges, textures, contours, and any other characteristics that describe a subject’s heart) extracted from plurality of ultrasonic images 124 using a convolutional neural network as described herein. Such parameterization may involve processor 104 to derive one or more shape parameters including one or more morphological descriptors that quantitatively describe a heart based on extracted features. In some cases, processor 104 may be configured to use principal component analysis (PCA) to reduce the dimensionality of set of shape parameters, allowing processor 104 to focusing on the most informative shape parameters of a set of shape parameters in further processing steps as described below.

[0156] Still referring to FIG. 1, in a non-limiting example, set of shape parameters may be generated based on plurality of ultrasonic images 124 using a shape identification model. Generating a set of shape parameters may include receiving structure training data, wherein the structure training data may include a plurality of image sets as inputs correlated to a plurality of

[0157] 42 Attorney Docket No. 1518-193PCT1 shape parameter sets as outputs. In some cases, structure training data may be received from a Data store. For example, and without limitation, structure training data may be used to show each ultrasonic image may indicate a particular set of shape parameters. In some embodiments, structure training data may include historical ultrasonic images correlated with historical computed tomography scan data. Such a training dataset may be used to train shape identification model to generate a set of shape parameters representing a structure’s shape as a function of a set of ultrasonic images, which may be input into the model in order to receive, as an output, a set of shape parameters. Shape identification model may be trained, by processor 104, using structure training data. Additionally, structure training data may include previously input image sets and their corresponding shape parameter outputs. Shape identification model may be iterative such that outputs may be used as future inputs of shape identification model. This may allow the shape identification model to evolve. Processor 104 may be further configured to generate set of shape parameters as a function of plurality of ultrasonic images 124 using the trained shape identification model.

[0158] Still referring to FIG. 1, generating a set of shape parameters may include performing image processing / segmentation techniques, as described above, prior to implementation of shape identification model in order to optimize performance and runtime of processor 104 and training of model. For example, image segmentation may include normalization and standardization methods performed by computer vision model to ensure that pixel values in ultrasonic images are normalized or standardized to a consistent scale thus aiding convergence during training of shape identification model. Image segmentation may include data augmentation techniques such as rotation, scaling, flipping, and translation to artificially increase the size of the training dataset and improve model generalization. Image segmentation may include image enhancement preprocessing techniques like histogram equalization or contrast stretching to enhance relevant features in the images. Image segmentation may include texture and shape descriptors to extract features beyond pixel values, such as texture and shape descriptors, to capture additional information about structures. Image segmentation may include architecture selection methods, as in experiments with different architectures, such as U-Net, DeepLab, or custom architectures, depending on the complexity and characteristics of the images. Image segmentation may include grid Search or random Search processing methods to systematically explore hyperparameter combinations to find the optimal configuration for a 3D model. As previously disclosed, image

[0159] 43 Attorney Docket No. 1518-193PCT1 segmentation may include separating specific structures or regions of interest (ROI) from the background or other structures in a given ultrasonic image, wherein a collection of ROIs may be also incorporated by the shape parameter training data / structure training data.

[0160] Still referring to FIG. 1, processor 104 may use a statistical shape model to generate and / or iteratively refine a 3D cardiac model 128 based on a set of shape parameters. In some cases, 3D cardiac model 128 may be generated through a direct 3D reconstruction from a series of (2D) ultrasonic images. In a non-limiting example, plurality of ultrasonic images 124 may include a plurality of ultrasonic images captured from different angles and positions within and / or around a structure. Processor 104 may be configured to apply one or more 3D reconstruction algorithms, such as without limitation, marching cubes, contour detection and segmentation, active contour models, and / or the like to create a coherent 3D representation e.g., 3D cardiac model 128 of a heart. In some cases, such direct 3D reconstruction may leverage the inherent spatial information within plurality of ultrasonic images 124, providing a direct and intuitive way to model 3D cardiac model 128. In a further embodiment, 3D modeling techniques may be applied to create the initial 3D model, such as surface modeling, solid modeling, or parametric modeling, among others. As an ordinary person skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various 3D reconstruction algorithms that may be used by processor 104 to generate 3D cardiac model 128. As used in this disclosure, a “statistical shape model” (SSM) is a data structure representing, including, and / or utilizing a mathematical model that captures principal modes of variation in shape across a population of structures. In some cases, SSM may be constructed by analyzing one or more datasets of shapes and identifying, for example, mean shapes and main modes of variation within the one or more datasets. In a non-limiting example, SSM may start with calculation of at least one mean shape, which represents an average geometry of all shapes of a structure in a given dataset, wherein the at least one mean shape may be served as a central reference point for processor 104 to understand different variations. In some embodiments, unique SSMs are created for different structure categories, such as different organs or tissues. In a non-limiting example, a first SSM may be created for a first structure category such as kidneys and a second SSM may be created for a second structure category such as hearts. In some cases, dataset may include, without limitation, structure training data, and / or datasets within ultrasonic image databases described herein. SSM may also identify one or more principal modes of variation within given datasets

[0161] 44 Attorney Docket No. 1518-193PCT1 described herein, wherein the “principal modes of variations,” for the purpose of this disclosure, refer to main patterns or directions along which data points vary within dataset. In a non-limiting example, identifying principal modes of variations may include applying principal component analysis (PCA) on given dataset. Additionally, or alternatively, shapes may be described directly using plurality of shape parameter sets (in structure training data). In some cases, shape parameter sets may correspond to a plurality of modes of variations. Further, one or more statistical constraints (e.g., mean, variance, correlation, boundary, proportion constraint and / or the like) may be introduced into SSM based on the distribution of shape parameters within plurality of shape parameter sets and / or 3D structure dimensions. In some embodiments, each shape parameter within a set of shape parameters may be associated with and / or comprise a corresponding parameter range. Such a parameter range may, for example, include a range of values associated with a normal and / or healthy heart. Such a parameter range may be determined based on, for example, a subset of possible values of a parameter which historical healthy structures commonly fall into, as determined from a dataset.

[0162] Still referring to FIG. 1, in some cases, once modes of variation are extracted, processor 104 may be configured to create a shape representation for any given structure shape within the studied class. In a non-limiting example, 3D cardiac model 128 having a shape S may be mathematically represented as S wherein S denotes the mean shape derived from the set of example shapes, M is the number of modes of variation considered, akare the coefficients or weights for each mode, and <pkare the modes of variation (eigenvectors corresponding to the kth principal component). In some cases, coefficients akmay dictate a degree to which each mode of variation is present in shape S. In some cases, coefficients akmay vary from positive to negative (or negative to positive) based on the deformation of the 3D cardiac model 128 in directions described by each mode of variation. In some cases, 3D cardiac model 128 may include mean shape as described herein. In some cases, 3D cardiac model 128 may include a predictive structure shape that may not have been explicitly seen in the set of example shapes or patient’s heart observations. In some cases, 3D cardiac model 128 may be in 3D VOR as described above.

[0163] Still referring to FIG. 1, generating the 3D cardiac model 128 may include transforming 3D cardiac model 128 to a second model as a function of a plurality of mode changers within SSM, wherein each mode changer of the plurality of mode changers is associated with a model

[0164] 45 Attorney Docket No. 1518-193PCT1 feature of 3D cardiac model 128. As used in this disclosure, a “mode changer” is an algorithmic component derived from PCA configured to encapsulate a specific mode of variation as described above (representing a distinct way in which the shape of 3D cardiac model 128 may deviate from the mean shape). A “model feature,” for the purpose of this disclosure, is a distinct, recognizable and quantifiable attribute or characteristic of the 3D cardiac model 128. For example, and without limitation, model feature may include an anatomical feature such as the size and curvature of the ventricles, the thickness of the heart wall, the positioning of heart valves or the like. In some cases, model feature may correspond to at least one shape parameter as described herein. In a non-limiting example, a mode changer may be associated with the size variation of the left ventricle identified within 3D cardiac model 128. Such mode changer may be adjusted to modify the volume of the left ventricle, resulting in a second 3D model that mimics potential biological variations or specific patient conditions that is different from original 3D cardiac model 128. In some cases, multiple mode changers of SSM may be adjusted simultaneously. For example, and without limitation, the rigid registration might involve translations and rotations to superimpose the shapes; affine registration could incorporate scaling, shearing, and other linear transformations; while non-rigid methods might employ B- splines, thin-plate splines, or diffeomorphic transformations to flexibly map one shape onto another. In some cases, an averaged position of each corresponding point (or voxel) across all example shapes may be calculated using formula Pjt, where pLis the mean position of the ith point (or voxel), Pjt is the position of the ith point in the / th example shape, and N is the total number of example shapes in the labeled set. In some cases, principle component analysis (PCA) may be applied to the aligned shapes to extract at least a primary mode of variation. As described herein, a “primary mode of variation” is a mode of variation that have the most significant variability, wherein the “mode of variation,” for the purpose of this disclosure, is a specific pattern or direction of a shape change. In some cases, such significancy may be indicated by the first principal component in PCA. In some cases, a plurality of modes of variation may be extracted, wherein each mode (or principal component) may represent a specific way the shape of structure may be deformed from the mean shape, determined by one or more eigenvectors of the covariance matrix of the aligned shapes. In a non-limiting example, eigenvector with the highest eigenvalue may represent primary mode of variation which captures the largest amount of shape variability within example shapes, while subsequent modes

[0165] 46 Attorney Docket No. 1518-193PCT1 (eigenvectors) capture decreasing amounts of variability. Still referring to FIG. 1, in some embodiments, apparatus 100 may determine cardiac measurement 148 based on 3D cardiac model 128. As used herein, a “cardiac measurement” is a quantitative value representing a physical characteristic of a heart. In some embodiments, a cardiac measurement may include an ostial diameter. In some embodiments, a cardiac measurement may include an ostial length. In some embodiments, a cardiac implant size 152 may be determined based on a cardiac measurement 148, such as a maximum ostial size. In some embodiments, a cardiac implant size 152 may be determined based on 3D cardiac model 128. A “cardiac implant size,” for the purposes of this disclosure, is a datum or data representing at least a dimension of a cardiac implant. A cardiac implant size may include a left atrial appendage occlusion device size. A cardiac implant size may include a Watchman device size.

[0166] Still referring to FIG.1, In some embodiments, a cardiac implant placement 156 may be determined based on a cardiac measurement 148 and / or 3D cardiac model 128. As used herein, a “cardiac implant placement” is a location of a cardiac implant, a location of a component of a cardiac implant, or both. In some embodiments, a cardiac implant placement may include a placement of a Watchman device. In some embodiments, a cardiac implant candidate quality 160 may be determined based on a cardiac measurement 148 and / or 3D cardiac model 128. As used herein, a “cardiac implant candidate quality” is a measure of whether a subject is a suitable candidate for receiving a cardiac implant, the degree to which a subject is a suitable candidate for receiving a cardiac implant, or both. In some embodiments, a cardiac implant candidate quality 160 may include a measure of whether a subject is a suitable candidate for receiving a Watchman device and / or a measure of the degree to which a subject is a suitable candidate for receiving a Watchman device. In some embodiments, a candidate’s suitability may be determined based on whether a thrombus is detected. As used herein, a “thrombus” is a blood clot formed in situ within the vascular system of a subject, where the blood clot impedes blood flow. In some embodiments, a thrombus machine learning model may be used to detect a thrombus based on plurality of ultrasonic images 124. In some embodiments, a thrombus machine learning model may detect one or more features within one or more images of plurality of ultrasonic images 124 and may determine a presence and / or absence of a thrombus based on such features. Such features may include, as non-limiting examples, irregularity of edges, shapes of structures detected within an image, mean intensity and / or standard deviation of intensity of a region. In

[0167] 47 Attorney Docket No. 1518-193PCT1 some embodiments, thrombus machine-learning model may be trained on training data correlating ultrasonic images to labeled ultrasonic images. In some embodiments, labeled ultrasonic images may indicate the presence or absence of a thrombus. In some embodiments, labeled ultrasonic images may indicate a location of a thrombus. For example, irregularity of edges may indicate that a structure includes a thrombus, which may have fewer regular edges than healthy vessel walls. In another example, a shape of a structure and / or a degree to which a structure has a circular or irregular shape may be used to distinguish a thrombus (which may have an irregular shape) from a healthy blood vessel (which may have a circular cross section).

[0168] Still referring to FIG. 1, apparatus 100 may include a display device. As used in this disclosure, a “display device” is an electronic device that visually presents information to a user. In an embodiment, display device may include a user interface that translates data such as, without limitation, cardiac measurement 148, 3D cardiac model 128, cardiac implant size 152, cardiac implant placement 156, and cardiac implant candidate quality 160. A user interface may include a graphical user interface (GUI), wherein the GUI may include a window in which data described herein may be displayed. In an embodiment, a user interface may include one or more graphical locator and / or cursor facilities allowing user to interact with cardiac measurement 148, 3D cardiac model 128, cardiac implant size 152, cardiac implant placement 156, and cardiac implant candidate quality 160, for instance, by using a touchscreen, touchpad, mouse, keyboard, and / or other manual data entry device. In some cases, 3D cardiac model 128 may be interactive. For instance, medical professionals may rotate, zoom, and / or explore 3D cardiac model 128 from various angles. A user may enter user input containing selecting specific regions, adding comments, adjusting parameter, and the like. In a non-limiting example, user interface may include one or more menus and / or panels permitting selection of measurements, models, visualization of data / model to be displayed and / or used, elements of data, functions, or other aspects of data / model to be edited, added, and / or manipulated, options for importation of and / or linking to application programmer interfaces (APIs), exterior services, data source, machinelearning models, and / or algorithms, or the like. Persons skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various ways in which a visual interface and / or elements thereof may be implemented and / or used as described in this disclosure.

[0169] Still referring to FIG. 1, in some embodiments, a computing device may determine a Left Atrial Appendage Occlusion Device placement as a function of 3D cardiac model 128. In some

[0170] 48 Attorney Docket No. 1518-193PCT1 embodiments, a computing device may determine a size of a Left Atrial Appendage Occlusion Device placement as a function of 3D cardiac model 128. In some embodiments, a computing device may determine whether there is leakage resulting from Left Atrial Appendage Occlusion Device placement as a function of 3D cardiac model 128. In some embodiments, a determined Left Atrial Appendage Occlusion Device size, placement, and / or leakage may be displayed to a user, such as by a display device.

[0171] Still referring to FIG. 1, in some embodiments, an apparatus and / or method described herein may allow ultrasonic imaging to replace and / or be an alternative to MRIs and / or CT scans. This may limit radiation exposure of subjects.

[0172] Still referring to FIG. 1, in some embodiments, use of a TEE sweep from multiple angles to recreate a 3D mesh of the left atrium anatomy may enable accurate measurement of maximal ostium diameters of the left atrial appendage (LAA) as needed for device sizing. This can help increase initial device size selection without necessitating any additional scanning or radiation exposure for patients. Improved device size selection may also result in lower number of devices per procedure and overall shorter procedure time. An apparatus or method described herein can further be used to guide transseptal puncture and navigation of a sheath and delivery system, helping ease the implantation procedure itself.

[0173] Still referring to FIG. 1, measurement of the sizes from ICE views may be more subjective and not as clear as with TEE. This may increase the cognitive burden on the physician to arrive at the best view where the ‘maximal’ ostial diameter is visible while navigating the catheter across the septum and within the left atrium. Additionally, a TEE probe may be manipulated and measured by a sonographer, freeing up a physician to focus on appropriate delivery and placement.

[0174] Still referring to FIG. 1, 3D reconstruction of the LAA from ICE may aid workflows of physicians who wish to perform an implantation procedure without TEE, making it easier to obtain the ‘maximal’ measurements on the reconstructed 3D LA / LAA mesh instead of navigating an ICE catheter to the appropriate view.

[0175] Now referring to FIG. 2, an exemplary embodiment of an ultrasonic image such as ICE image 200 is illustrated. As described above with reference to FIG. 1, plurality of ultrasonic images 124 may include a plurality of ICE images, wherein each ICE image of the plurality of ICE images is a specialized form of echocardiography that may provide detailed image of heart’s

[0176] 49 Attorney Docket No. 1518-193PCT1 interior structures. In a non-limiting example, plurality of ICE images may include an ICE video (e.g., plurality of ICE images arranged in a corresponding time sequence). In an embodiment, ICE image 200 may be real-time, dynamic ultrasound image that provide a (detailed) view 204 of heart’s interior structures, including, without limitation, right atrium (RA) 208, anterior descending (AD) 212, pulmonary atresia (PA) 216, and right ventricular (RV) 220.

[0177] With continued reference to FIG. 2, in some cases, ICE image 200 may include gray scaled image. It should be noted that, in some cases, ICE image 200 may be configured to visualize blood flow and / or blood flow patterns within the heart via color doppler. In some cases, resolution and / or clarity of ICE image 200 as described herein may be superior to transthoracic or transesophageal echocardiography due to the ICE catheter may be positioned inside the heart, closer to the structures being imaged.

[0178] Still referring to FIG. 2, in a non-limiting example, heart chambers may appear as dark, anechoic (black) areas since they are filled with blood, which does not reflect ultrasound waves well. Heart walls, valves, and / or other structures may appear as varying shades of gray, depending on their density and composition, in some cases, Color Doppler overlays may show blood flow in different colors, indicating the direction and speed of blood flow. For instance, and without limitation, red may indicate flow towards the probe, while blue may indicate flow away from the probe.

[0179] With continued reference to FIG. 2, in a non-limiting embodiment, ICE image 200 may be synchronized with ECG data, allowing for precise timing of cardiac events with anatomical visualization provided by ICE. In some cases, ICE image 200 may include an ECG display 224 configured to display ECG waveform as a continuous line graph at the top, bottom, or side of ICE image 200. In some cases, specific parts of the cardiac cycle e.g., systole or diastole, may be correlated with visual data from ICE image 200.

[0180] Additionally, or alternatively, and still referring to FIG. 2, ICE image 200 may come with accompanying metadata 228 displayed on the side or comers of ICE image 200 as described herein. In some cases, metadata 228 may provide essential contextual information about ICE image 200 and / or the corresponding patient. In a non-limiting example, metadata 228 may include patient information (e.g., patient ID, name, DOB, age, gender, and the like), image acquisition details (e.g., date and time, probe type, frequency, depth, gain, and the like), procedure-related information (e.g., procedure name, operator, location, and the like), ECG trace

[0181] 50 Attorney Docket No. 1518-193PCT1 (e g., ECG data as described above), measurement annotations (e.g., any measurements taken directly on the image e.g., diameter, a value of thickness of a heart wall and the like), image sequence information (e.g., image number, total number of frames, and the like), comments or notes, hospital or clinic information, and / or the like. As an ordinary person skilled in the art, upon reviewing the entirety of this disclosure, will be aware of ICE image 200 and various components thereof may be incorporated by apparatus 100 for generating 3D model of cardiac anatomy.

[0182] Now referring to FIG. 3, a flow diagram of an exemplary embodiment of an ultrasonic image such as ICE example generation process 300. In an embodiment, structure training data may be generated, at least in part, via ICE example generation process 300. In some cases, processor 104 may be configured to receive a 3D model of the heart, such as any 3D model of heart 320 as described herein and identify an ICE view 304 (i.e., visual representation of image obtained using intracardiac echocardiography as described above e.g., ICE image 200) based on the received 3D model. In some cases, 3D model received by processor 104 may be derived from CT scans as described above with reference to FIG. 1. In other cases, processor 104 may receive CT scans directly instead of 3D models. A synthetic ICE frame 308 may then be generated, by processor 104, as a function of identified ICE view 304, wherein the synthetic ICE frame 308 may be used as one or the training examples in structure training data.

[0183] With continued reference to FIG. 3, in some cases, processor 104 may interface with one or more 3D models (i.e., detailed representation of heart’s anatomy in a 3D space, capturing intricate structures, chambers, vessels, valves, among others) as described above, or other imaging modalities and / or databases, and equipped with algorithms e.g., CNN, gradient boosting machines, SVM, PC A, and / or the like to analyze model’s geometry and spatial relationships upon receiving the 3D models. In some cases, 3D models may be received from SSM 324 as described above with reference to FIG. 1 via a communicative connection between processor 104 and SSM 324. In a non-limiting example, processor 104 may be configured to determine an optimal viewpoints or angles from which ICE view 304 would provide a desired diagnostic value or procedural guidance.

[0184] Still referring to FIG. 3, in some cases, identification and selection of ICE view 304 may be automatically identified, using one or more machine learning models as described herein. In a non-limiting example, processor 104 may utilize one or more machine learning models trained

[0185] 51 Attorney Docket No. 1518-193PCT1 on cardiac anatomy viewpoints identification training data, wherein the cardiac anatomy viewpoints identification training data may include a plurality of cardiac anatomies as input correlated to a plurality of ICE images as output and identify at least one ICE view 304 (most informative) for a given cardiac anatomy using the trained machine learning models.

[0186] Still referring to FIG. 3, in other cases, ICE view 304 may be defined by a user such as a medical professional. In a non-limiting example user interface of display device may allow a user (e.g., a clinician) to manually rotate, pan, and zoom displayed 3D model and / or corresponding CT scans. As user do so, processor 104 may dynamically calculate and displays potential ICE views 304 based on user’s chosen perspective. Additionally, or alternatively, depending on cardiac procedure being planned or executed, processor 104 may prioritize certain ICE views 304. For instance, and without limitation, ICE view 304 may be pre-defined. For atrial fibrillation ablation, ICE view 304 may showcase the pulmonary veins’ entrances into the LA may be emphasized. In other cases, ICE view 304 may be automatically identified, by processor 104, using one or more machine learning models as described herein, such as, without limitation, synthetic ICE data generator as described in detail below.

[0187] With continued reference to FIG. 3, as used in this disclosure, a “synthetic ICE frame” refers to a digitally generated or simulated image that emulates a visual representation obtained from ICE view 304. In some cases, synthetic ICE frames 308 may be produced using computational methods and / or models such as, without limitation, a synthetic ICE data generator 312 based on pre-existing data, models, or simulations e.g., identified ICE views 304. In a nonlimiting example, synthetic ICE frames 308 may include a simplified version e.g., an image illustrating heart anatomy via a plurality of lines indicating contours of heart’s structure as shown in FIG. 3. One or more image processing techniques and / or computer vision algorithms such as, without limitation, histogram equalization, adaptive filtering, edge detection (e.g., Canny or Sobel operators), contour extraction, and / or the like may be applied, by processor 104, on a segmented CT scan and / or 3D models based on identified ICE view 304. Synthetic ICE frame 308 may be rendered on a blank canvas or background that mimics the echogenicity of an ICE image according to extracted contours, wherein the extracted contours may be represented as a bold lines and enhanced with shading to give depth. In some cases, synthetic ICE frame 308 may be validated and verified by overlaying synthetic ICE frame 308 onto original ICE view 304, ensuring accuracy and resemblance.

[0188] 52 Attorney Docket No. 1518-193PCT1 Still referring to FIG. 3, in some cases, generating synthetic ICE frames 308 may include implementations of one or more aspects of “generative artificial intelligence,” a type of Al that uses machine learning algorithms to create, establish, or otherwise generate data. Such data may include, without limitation, ultrasonic image that is similar to one or more provided training examples. In an embodiment, machine learning module described herein may generate one or more generative machine learning models that are trained on one or more set of CT scans and / or 3D models in ICE image view 304 as described above. Synthetic ICE data generator 312 may include one or more generative machine learning models may be configured to generate new examples that are similar to the training data of the one or more generative machine learning models but are not exact replicas; for instance, and without limitation, data quality or attributes of the generated examples may bear a resemblance to the training data provided to one or more generative machine learning models, wherein the resemblance may pertain to underlying patterns, features, or structures found within the provided training data.

[0189] Still referring to FIG. 3, in some cases, generative machine learning models within synthetic ICE data generator may include one or more generative models. As described herein, “generative models” refers to statistical models of the joint probability distribution P(X, E) on a given observable variable x, representing features or data that can be directly measured or observed (e.g. CT scans and / or 3D models derived from CT scans) and target variable y, representing the outcomes or labels that one or more generative models aims to predict or generate (e g., synthetic ICE frames 308). In some cases, generative models may rely on Bayes theorem to find joint probability; for instance, and without limitation, Naive Bayes classifiers may be employed by computing device to categorize input data such as, without limitation, CT scans and / or 3D models derived from CT scans into different views.

[0190] In a non-limiting example, and still referring to FIG. 3, one or more generative machine learning models may include one or more Naive Bayes classifiers generated, by processor 104, using a Naive bayes classification algorithm. Naive Bayes classification algorithm generates classifiers by assigning class labels to problem instances, represented as vectors of element values. Class labels are drawn from a finite set. Naive Bayes classification algorithm may include generating a family of algorithms that assume that the value of a particular element is independent of the value of any other element, given a class variable. Naive Bayes classification algorithm may be based on Bayes Theorem expressed as P(A / B)= P(B / A) P(A)-^P(B), where

[0191] 53 Attorney Docket No. 1518-193PCT1 P(A / B) is the probability of hypothesis A given data B also known as posterior probability; P(B / A) is the probability of data B given that the hypothesis A was true; P(A) is the probability of hypothesis A being true regardless of data also known as prior probability of A; and P(B) is the probability of the data regardless of the hypothesis. A naive Bayes algorithm may be generated by first transforming training data into a frequency table. Processor 104 may then calculate a likelihood table by calculating probabilities of different data entries and classification labels. Processor 104 may utilize a naive Bayes equation to calculate a posterior probability for each class. A class containing the highest posterior probability is the outcome of prediction.

[0192] Still referring to FIG. 3, although Naive Bayes classifier may be primarily known as a probabilistic classification algorithm; however, it may also be considered a generative model described herein due to its capability of modeling the joint probability distribution P(X, K) over observable variables X and target variable Y. In an embodiment, Naive Bayes classifier may be configured to make an assumption that the features X are conditionally independent given class label Y, allowing generative model to estimate the joint distribution as P(X, K) = P(Y) [iP(Xi I F), wherein P(F) may be the prior probability of the class, and P(X |F) is the conditional probability of each feature given the class. One or more generative machine learning models containing Naive Bayes classifiers may be trained on labeled training data, estimating conditional probabilities P(X(|F) and prior probabilities P(F) for each class; for instance, and without limitation, using techniques such as Maximum Likelihood Estimation (MLE). One or more generative machine learning models containing Naive Bayes classifiers may select a class label y according to prior distribution P(F), and for each feature Xt. sample at least a value according to conditional distribution P(X y). Sampled feature values may then be combined to form one or more new data instance with selected class label y. In a non-limiting example, one or more generative machine learning models may include one or more Naive Bayes classifiers to generate new examples of ICE images based on CT scans and / or 3D models derived from CT scans (e.g., identified ICE views 304), wherein the models may be trained using training data containing a plurality of features of input data as described herein and / or the like correlated to a plurality of ICE views.

[0193] Still referring to FIG. 3, in some cases, one or more generative machine learning models may include generative adversarial network (GAN). As used in this disclosure, a “generative adversarial network” is a type of artificial neural network with at least two sub models (e.g.,

[0194] 54 Attorney Docket No. 1518-193PCT1 neural networks), a generator, and a discriminator, that compete against each other in a process that ultimately results in the generator learning to generate new data samples, wherein the “generator” is a component of the GAN that learns to create hypothetical data by incorporating feedbacks from the “discriminator” configured to distinguish real data from the hypothetical data. In some cases, generator may learn to make discriminator classify its output as real. In an embodiment, discriminator may include a supervised machine learning model while generator may include an unsupervised machine learning model as described in further detail with reference to FIGS. 6-8.

[0195] With continued reference to FIG. 3, in an embodiment, discriminator may include one or more discriminative models, i.e., models of conditional probability P(E|X = x) of target variable Y, given observed variable X. In an embodiment, discriminative models may learn boundaries between classes or labels in given training data. In a non-limiting example, discriminator may include one or more classifiers as described in further detail below with reference to FIG. 6 to distinguish between different categories e.g., real vs. fake, or states e.g., TRUE vs. FALSE within the context of generated data such as, without limitations, synthetic ICE frames 308, and / or the like. In some cases, processor 104 may implement one or more classification algorithms such as, without limitation, Support Vector Machines (SVM), Logistic Regression, Decision Trees, and / or the like to define decision boundaries.

[0196] In a non-limiting example, and still referring to FIG. 3, generator of GAN may be responsible for creating synthetic data that resembles real ICE images. In some cases, GAN may be configured to receive CT scans and / or 3D models derived from CT scans as input and generates corresponding examples of ICE images containing information describing anatomy in different ICE views. On the other hand, discriminator of GAN may evaluate the authenticity of the generated content by comparing it to true ICE images, for example, discriminator may distinguish between genuine and generated content and providing feedback to generator to improve the model performance. Additionally, or alternatively, GAN may include a conditional GAN as an extension of the basic GAN as described herein that allows for generation of ICE images using pre-existing CT scans and / or 3D models derived from CT scans based on certain conditions or labels. In standard GAN, generator may produce samples from random noise, while in a conditional GAN, generator may produce samples based on random noise and a given condition or label.

[0197] 55 Attorney Docket No. 1518-193PCT1 With continued reference to FIG. 3, in other embodiments, one or more generative models may also include a variational autoencoder (VAE). As used in this disclosure, a “variational autoencoder” is an autoencoder (i.e., an artificial neural network architecture) whose encoding distribution is regularized during the model training process in order to ensure that its latent space includes desired properties allowing new data sample generation. In an embodiment, VAE may include a prior and noise distribution respectively, trained using expectationmaximization meta-algorithms such as, without limitation, probabilistic PC A, sparse coding, among others. In a non-limiting example, VEA may use a neural network as an amortized approach to jointly optimize across input data and output a plurality of parameters for corresponding variational distribution as it maps from a known input space to a low-dimensional latent space. Additionally, or alternatively, VAE may include a second neural network, for example, and without limitation, a decoder, wherein the “decoder” is configured to map from the latent space to the input space.

[0198] In a non-limiting example, and still referring to FIG. 3, VAE may be used by processor 104 to model complex relationships between CT scans and / or 3D models derived from CT scans. In some cases, VAE may encode input data into a latent space, capturing example ICE images. Such encoding process may include learning one or more probabilistic mappings from observed CT scans and / or 3D models derived from CT scans to a lower-dimensional latent representation. Latent representation may then be decoded back into the original data space, therefore reconstructing the 3D models representing example ICE images. In some cases, such decoding process may allow VAE to generate new examples or variations that are consistent with the learned distributions.

[0199] Additionally, or alternatively, and still referring to FIG. 3, processor 104 may be configured to continuously monitor synthetic ICE data generator. In an embodiment, processor 104 may configure discriminator to provide ongoing feedback and further corrections as needed to subsequent input data. An iterative feedback loop may be created as processor 104 continuously receive real-time data, identify errors (e.g., distance between synthetic ICE frame 308 and real ICE images) as a function of real-time data, delivering corrections based on the identified errors, and monitoring subsequent model outputs and / or user feedbacks on the delivered corrections. In an embodiment, processor 104 may be configured to retrain one or more generative machine learning models within synthetic ICE data generator based on user modified

[0200] 56 Attorney Docket No. 1518-193PCT1 ICE frames or update training data of one or more generative machine learning models within synthetic ICE data generator by integrating validated synthetic ICE frames (i.e., subsequent model output) into the original training data. In such embodiment, iterative feedback loop may allow synthetic ICE data generator to adapt to the user’s needs and performance requirements, enabling one or more generative machine learning models described herein to learn and update based on user responses and generated feedbacks.

[0201] With continued reference to FIG. 3, other exemplary embodiments of generative machine learning models may include, without limitation, long short-term memory networks (LSTMs), (generative pre-trained) transformer (GPT) models, mixture density networks (MDN), and / or the like. As an ordinary person skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various generative machine learning models that may be used generating synthetic ICE frames 308.

[0202] Still referring to FIG. 3, in a further non-limiting embodiment, synthetic ICE data generator 312 may be further configured to generate a multi-model neural network that combines various neural network architectures described herein. In a non-limiting example, multi-model neural network may combine LSTM for time-series analysis with GPT models for natural language processing. Such fusion may be applied by computing device to generate synthetic ICE frames 308. In some cases, multi-model neural network may also include a hierarchical multimodel neural network, wherein the hierarchical multi-model neural network may involve a plurality of layers of integration; for instance, and without limitation, different models may be combined at various stages of the network. Convolutional neural network (CNN) may be used for image feature extraction, followed by LSTMs for sequential pattern recognition, and a MDN at the end for probabilistic modeling. Other exemplary embodiments of multi-model neural network may include, without limitation, ensemble-based multi-model neural network, cross- modal fusion, adaptive multi-model network, among others. As an ordinary person skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various generative machine learning models that may be used to generating synthetic ICE frames 308 as described herein. As an ordinary person skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various multi-model neural network and combination thereof that may be implemented by apparatus 100 in consistent with this disclosure.

[0203] Now referring to FIG. 4, an exemplary embodiment of a 3D VOR 400 is illustrated. 3D

[0204] 57 Attorney Docket No. 1518-193PCT1 VOR 400 may be used to represent 3D object 404. In an embodiment, 3D VOR 400 may divide a 3D space 408 into a grid of one or more cubic units e.g., voxels 412, wherein each voxel 412 represents a specific volume within 3D space 408. In a non-limiting example, 3D object 404 may include a structure pertaining to a subject.

[0205] Still referring to FIG. 4, in some cases, each voxel 412 may act as a basic building block. In a non-limiting example, each voxel 412 may be configured to represent a discrete portion of 3D space 408. In an embodiment, each voxel 412 may include a presence indicator as described above with reference to FIG. 1, which denotes whether the voxel is occupied or unoccupied. In such embodiment, the binary or continuous value may allow 3D VOR 400 to map the presence or absence of material within each voxel 412, creating a granular representation of 3D object 404.

[0206] With continued reference to FIG. 4, in some cases, the resolution of 3D VOR 400 may be determined by the size and number of voxels within the grid. In a non-limiting example, smaller voxel may provide a higher resolution, capturing finer details, while larger voxels offer a more generalized representation.

[0207] Still referring to FIG. 4, in an embodiment, voxels 412 may be arranged in a regular pattern along three axis 416a-c, each pointing a distinct direction. In a non-limiting example, voxels 412 may be arranged along x, y, and z axes, wherein such arrangement may facilitate efficient manipulation and rendering of the 3D object 404. In some cases, spatial features 420a-c such as, without limitation, edges, surfaces, textures, and any other spatial features as described above with reference to FIG. 1, may be extracted from 3D VOR 400 by analyzing the relationships and patterns between neighboring voxels.

[0208] Now referring to FIG. 5, a schematic of an exemplary transesophageal echocardiogram (TEE) procedure 500 is shown. In some cases, TEE 500 may be performed during another procedure for instance heart surgery. According to some embodiments, a patient 504 has an endoscope 508, with an ultrasonic transducer 512, inserted into his esophagus 516. As one’s esophagus 516 is proximal one’s heart 520, ultrasonic transducer 512 may generate echocardiograms.

[0209] Still referring to FIG. 5, in some embodiments, transesophageal echocardiography (TEE) may provide superior imaging quality than intracardiac echocardiography (ICE), as larger ultrasound transducers 512 may be placed within the esophagus 516 than within heart 520. In

[0210] 58 Attorney Docket No. 1518-193PCT1 some cases, ultrasound transducers must be substantially miniaturized to fit within heart 520, as in ICE catheters. As esophagus 516 may be proximal to heart 520, TEE may provide a clear image of various heart structures without needing vascular access (as commonly required by ICE). Additionally, TEE may be performed without obstructing patient’s 504 ribcage and intermediary tissues (as commonly required by transthoracic echocardiography [TTE]). In some cases, TEE images may also provide information associated with angle of acquisition. Angle of acquisition may be an angle of TEE probe with respect to esophagus 516 (e.g., esophageal axis).

[0211] Still referring to FIG. 5, in some embodiments, TEE echocardiogram data, including images showing heart structures and, in some cases, angle of acquisition, may be used as input to any machine learning process described in this application, for instance with reference to FIGS. 6-8. For instance TEE echocardiogram data may be used to reconstruct 3D heart models. In some cases, TEE echocardiogram data is input into a machine learning model that outputs a 3D heart model (e.g., 3D mesh model and / or statistical shape model).

[0212] Still referring to FIG. 5, in some embodiments, TEE may be a preferred imaging modality for structural heart interventions, such as without limitation left atrial appendage occlusion (LAOO) and aortic / mitral / other heart valve replacement procedures. In some cases, technology and improvements described in this disclosure permit creation and / or modification of a 3D heart mesh from TEE data to aid in planning implant size selection, as well as to guide implantation procedures. In some cases, virtual placement of a 3D model of a candidate implant (such as without limitation LAAO device and / or heart valve implants) can be simulated on a 3D heart model generated by any method described in this disclosure. This novel and improved functionality may validate appropriate size and placement of implants within heart 520, as well as other organs within body of patient 504. For example, in the context of electrophysiology procedures, TEE 500 can be used to create heart anatomical models that can be used as reference for electroanatomic mapping, and guidance of ablation catheters for atrial fibrillation procedures (such as without limitation pulmonary vein isolation).

[0213] Still referring to FIG. 5, in some embodiments, applications described with reference to TEE 500 above can be extended for use with TTE and point of care ultrasound (POCUS). In some cases, both TTE and POCUS may acquire ultrasound images of chest / surface of patient 504. In some cases, TTE and POCUS data may be used as an input (and / or training data) for any machine learning process described in this disclosure, for instance with reference to FIGS. 6-8.

[0214] 59 Attorney Docket No. 1518-193PCT1 In some cases, use of TTE and / or POCUS data (in machine learning processes described in this disclosure) may require adjustment in ultrasound acquisition parameters and positions to acquire a sufficient number of frames for 3D reconstruction. In some cases, TTE and POCUS offer improved accessibility (with POCUS being portable / mobile as well) and non-invasive 3D heart modeling, often without anesthesia or sedation, compared to catheterized 3D heart modeling commonly performed today for electroanatomical mapping and ablation procedures.

[0215] Referring now to FIG. 6, multiple 2D transesophageal echocardiogram (TEE) views at varying orientations are presented. Image A demonstrates a maximal LAA diameter of 21.1mm obtained utilizing the double-oblique multiplanar reconstruction methodology by 3D CT. Panels B-E demonstrate the traditional 2D TEE LAA scanning views of 0°, 45°, 90°, and 135° views reflected on a multiplanar reconstruction of the LAA ostium on 3D CT. In panels F-I, the CT dimensions are taken at a focal intersection point and angled, from 0° (assigned as the measurement of 17.1mm), and increased by 45° around this focal point for each subsequent measurement. Panel I demonstrates in this patient the 135° TEE measurement of 19.4mm as off- axis to the true centroid of the LAA, and is not reflective of the maximal LAA width as shown in panel A, thereby resulting in an undersized WATCHMAN device. By CT, this patient would receive a 24mm device with no peri-watchman leak.

[0216] Referring now to FIG. 7, an exemplary embodiment of a machine-learning module 700 that may perform one or more machine-learning processes as described in this disclosure is illustrated. Machine-learning module may perform determinations, classification, and / or analysis steps, methods, processes, or the like as described in this disclosure using machine learning processes. A “machine learning process,” as used in this disclosure, is a process that automatedly uses training data 704 to generate an algorithm instantiated in hardware or software logic, data structures, and / or functions that will be performed by a computing device / module to produce outputs 708 given data provided as inputs 712; this is in contrast to a non-machine learning software program where the commands to be executed are determined in advance by a user and written in a programming language.

[0217] Still referring to FIG. 7, “training data,” as used herein, is data containing correlations that a machine-learning process may use to model relationships between two or more categories of data elements. For instance, and without limitation, training data 704 may include a plurality of data entries, also known as “training examples,” each entry representing a set of data elements

[0218] 60 Attorney Docket No. 1518-193PCT1 that were recorded, received, and / or generated together; data elements may be correlated by shared existence in a given data entry, by proximity in a given data entry, or the like. Multiple data entries in training data 704 may evince one or more trends in correlations between categories of data elements; for instance, and without limitation, a higher value of a first data element belonging to a first category of data element may tend to correlate to a higher value of a second data element belonging to a second category of data element, indicating a possible proportional or other mathematical relationship linking values belonging to the two categories. Multiple categories of data elements may be related in training data 704 according to various correlations; correlations may indicate causative and / or predictive links between categories of data elements, which may be modeled as relationships such as mathematical relationships by machine-learning processes as described in further detail below. Training data 704 may be formatted and / or organized by categories of data elements, for instance by associating data elements with one or more descriptors corresponding to categories of data elements. As a nonlimiting example, training data 704 may include data entered in standardized forms by persons or processes, such that entry of a given data element in a given field in a form may be mapped to one or more descriptors of categories. Elements in training data 704 may be linked to descriptors of categories by tags, tokens, or other data elements; for instance, and without limitation, training data 704 may be provided in fixed-length formats, formats linking positions of data to categories such as comma-separated value (CSV) formats and / or self-describing formats such as extensible markup language (XML), JavaScript Object Notation (JSON), or the like, enabling processes or devices to detect categories of data.

[0219] Alternatively or additionally, and continuing to refer to FIG. 7, training data 704 may include one or more elements that are not categorized; that is, training data 704 may not be formatted or contain descriptors for some elements of data. Machine-learning algorithms and / or other processes may sort training data 704 according to one or more categorizations using, for instance, natural language processing algorithms, tokenization, detection of correlated values in raw data and the like; categories may be generated using correlation and / or other processing algorithms. As a non-limiting example, in a corpus of text, phrases making up a number “n” of compound words, such as nouns modified by other nouns, may be identified according to a statistically significant prevalence of n-grams containing such words in a particular order; such an n-gram may be categorized as an element of language such as a “word” to be tracked

[0220] 61 Attorney Docket No. 1518-193PCT1 similarly to single words, generating a new category as a result of statistical analysis. Similarly, in a data entry including some textual data, a person’s name may be identified by reference to a list, dictionary, or other compendium of terms, permitting ad-hoc categorization by machinelearning algorithms, and / or automated association of data in the data entry with descriptors or into a given format. The ability to categorize data entries automatedly may enable the same training data 704 to be made applicable for two or more distinct machine-learning algorithms as described in further detail below. Training data 704 used by machine-learning module 700 may correlate any input data as described in this disclosure to any output data as described in this disclosure. As a non-limiting illustrative example inputs may include ultrasonic image data and outputs may include 3D cardiac models.

[0221] Further referring to FIG. 7, training data may be filtered, sorted, and / or selected using one or more supervised and / or unsupervised machine-learning processes and / or models as described in further detail below; such models may include without limitation a training data classifier 716. Training data classifier 716 may include a “classifier,” which as used in this disclosure is a machine-learning model as defined below, such as a data structure representing and / or using a mathematical model, neural net, or program generated by a machine learning algorithm known as a “classification algorithm,” as described in further detail below, that sorts inputs into categories or bins of data, outputting the categories or bins of data and / or labels associated therewith. A classifier may be configured to output at least a datum that labels or otherwise identifies a set of data that are clustered together, found to be close under a distance metric as described below, or the like. A distance metric may include any norm, such as, without limitation, a Pythagorean norm. Machine-learning module 700 may generate a classifier using a classification algorithm, defined as a processes whereby a computing device and / or any module and / or component operating thereon derives a classifier from training data 704. Classification may be performed using, without limitation, linear classifiers such as without limitation logistic regression and / or naive Bayes classifiers, nearest neighbor classifiers such as k-nearest neighbors classifiers, support vector machines, least squares support vector machines, fisher’s linear discriminant, quadratic classifiers, decision trees, boosted trees, random forest classifiers, learning vector quantization, and / or neural network-based classifiers. As a non-limiting example, training data classifier 716 may classify elements of training data to subject demographic categories.

[0222] 62 Attorney Docket No. 1518-193PCT1 Still referring to FIG. 7, computing device may be configured to generate a classifier using a Naive Bayes classification algorithm. Naive Bayes classification algorithm generates classifiers by assigning class labels to problem instances, represented as vectors of element values. Class labels are drawn from a finite set. Naive Bayes classification algorithm may include generating a family of algorithms that assume that the value of a particular element is independent of the value of any other element, given a class variable. Naive Bayes classification algorithm may be based on Bayes Theorem expressed as P(A / B)= P(B / A) P(A)-^P(B), where P(A / B) is the probability of hypothesis A given data B also known as posterior probability; P(B / A) is the probability of data B given that the hypothesis A was true; P(A) is the probability of hypothesis A being true regardless of data also known as prior probability of A; and P(B) is the probability of the data regardless of the hypothesis. A naive Bayes algorithm may be generated by first transforming training data into a frequency table. Computing device may then calculate a likelihood table by calculating probabilities of different data entries and classification labels. Computing device may utilize a naive Bayes equation to calculate a posterior probability for each class. A class containing the highest posterior probability is the outcome of prediction. Naive Bayes classification algorithm may include a gaussian model that follows a normal distribution. Naive Bayes classification algorithm may include a multinomial model that is used for discrete counts. Naive Bayes classification algorithm may include a Bernoulli model that may be utilized when vectors are binary.

[0223] With continued reference to FIG. 7, computing device may be configured to generate a classifier using a K-nearest neighbors (KNN) algorithm. A “K-nearest neighbors algorithm” as used in this disclosure, includes a classification method that utilizes feature similarity to analyze how closely out-of-sample- features resemble training data to classify input data to one or more clusters and / or categories of features as represented in training data; this may be performed by representing both training data and input data in vector forms, and using one or more measures of vector similarity to identify classifications within training data, and to determine a classification of input data. K-nearest neighbors algorithm may include specifying a K-value, or a number directing the classifier to select the k most similar entries training data to a given sample, determining the most common classifier of the entries in the database, and classifying the known sample; this may be performed recursively and / or iteratively to generate a classifier that may be used to classify input data as further samples. For instance, an initial set of samples

[0224] 63 Attorney Docket No. 1518-193PCT1 may be performed to cover an initial heuristic and / or “first guess” at an output and / or relationship, which may be seeded, without limitation, using expert input received according to any process as described herein. As a non-limiting example, an initial heuristic may include a ranking of associations between inputs and elements of training data. Heuristic may include selecting some number of highest-ranking associations and / or training data elements.

[0225] With continued reference to FIG. 7, generating k-nearest neighbors algorithm may generate a first vector output containing a data entry cluster, generating a second vector output containing an input data, and calculate the distance between the first vector output and the second vector output using any suitable norm such as cosine similarity, Euclidean distance measurement, or the like. Each vector output may be represented, without limitation, as an n- tuple of values, where n is at least two values. Each value of n-tuple of values may represent a measurement or other quantitative value associated with a given category of data, or attribute, examples of which are provided in further detail below; a vector may be represented, without limitation, in n-dimensional space using an axis per category of value represented in n-tuple of values, such that a vector has a geometric direction characterizing the relative quantities of attributes in the n-tuple as compared to each other. Two vectors may be considered equivalent where their directions, and / or the relative quantities of values within each vector as compared to each other, are the same; thus, as a non-limiting example, a vector represented as [5, 10, 15] may be treated as equivalent, for purposes of this disclosure, as a vector represented as [1, 2, 3], Vectors may be more similar where their directions are more similar, and more different where their directions are more divergent; however, vector similarity may alternatively or additionally be determined using averages of similarities between like attributes, or any other measure of similarity suitable for any n-tuple of values, or aggregation of numerical similarity measures for the purposes of loss functions as described in further detail below. Any vectors as described herein may be scaled, such that each vector represents each attribute along an equivalent scale of values. Each vector may be “normalized,” or divided by a “length” attribute, such as a length attribute I as derived using a Pythagorean norm: where at is attribute number i of the vector. Scaling and / or normalization may function to make vector comparison independent of absolute quantities of attributes, while preserving any dependency on similarity of attributes; this may, for instance, be advantageous where cases represented in training data are represented by different quantities of samples, which may result in proportionally equivalent vectors with

[0226] 64 Attorney Docket No. 1518-193PCT1 divergent values.

[0227] With further reference to FIG. 7, training examples for use as training data may be selected from a population of potential examples according to cohorts relevant to an analytical problem to be solved, a classification task, or the like. Alternatively or additionally, training data may be selected to span a set of likely circumstances or inputs for a machine-learning model and / or process to encounter when deployed. For instance, and without limitation, for each category of input data to a machine-learning process or model that may exist in a range of values in a population of phenomena such as images, user data, process data, physical data, or the like, a computing device, processor, and / or machine-learning model may select training examples representing each possible value on such a range and / or a representative sample of values on such a range. Selection of a representative sample may include selection of training examples in proportions matching a statistically determined and / or predicted distribution of such values according to relative frequency, such that, for instance, values encountered more frequently in a population of data so analyzed are represented by more training examples than values that are encountered less frequently. Alternatively or additionally, a set of training examples may be compared to a collection of representative values in a database and / or presented to a user, so that a process can detect, automatically or via user input, one or more values that are not included in the set of training examples. Computing device, processor, and / or module may automatically generate a missing training example; this may be done by receiving and / or retrieving a missing input and / or output value and correlating the missing input and / or output value with a corresponding output and / or input value collocated in a data record with the retrieved value, provided by a user and / or other device, or the like.

[0228] Continuing to refer to FIG. 7, computer, processor, and / or module may be configured to preprocess training data. “Preprocessing” training data, as used in this disclosure, is transforming training data from raw form to a format that can be used for training a machine learning model. Preprocessing may include sanitizing, feature selection, feature scaling, data augmentation and the like.

[0229] Still referring to FIG. 7, computer, processor, and / or module may be configured to sanitize training data. “Sanitizing” training data, as used in this disclosure, is a process whereby training examples are removed that interfere with convergence of a machine-learning model and / or process to a useful result. For instance, and without limitation, a training example may

[0230] 65 Attorney Docket No. 1518-193PCT1 include an input and / or output value that is an outlier from typically encountered values, such that a machine-learning algorithm using the training example will be adapted to an unlikely amount as an input and / or output; a value that is more than a threshold number of standard deviations away from an average, mean, or expected value, for instance, may be eliminated. Alternatively or additionally, one or more training examples may be identified as having poor quality data, where “poor quality” is defined as having a signal to noise ratio below a threshold value. Sanitizing may include steps such as removing duplicative or otherwise redundant data, interpolating missing data, correcting data errors, standardizing data, identifying outliers, and the like. In a nonlimiting example, sanitization may include utilizing algorithms for identifying duplicate entries or spell-check algorithms.

[0231] As a non-limiting example, and with further reference to FIG. 7, images used to train an image classifier or other machine-learning model and / or process that takes images as inputs or generates images as outputs may be rejected if image quality is below a threshold value. For instance, and without limitation, computing device, processor, and / or module may perform blur detection, and eliminate one or more Blur detection may be performed, as a non-limiting example, by taking Fourier transform, or an approximation such as a Fast Fourier Transform (FFT) of the image and analyzing a distribution of low and high frequencies in the resulting frequency-domain depiction of the image; numbers of high-frequency values below a threshold level may indicate blurriness. As a further non-limiting example, detection of blurriness may be performed by convolving an image, a channel of an image, or the like with a Laplacian kernel; this may generate a numerical score reflecting a number of rapid changes in intensity shown in the image, such that a high score indicates clarity and a low score indicates blurriness. Blurriness detection may be performed using a gradient-based operator, which measures operators based on the gradient or first derivative of an image, based on the hypothesis that rapid changes indicate sharp edges in the image, and thus are indicative of a lower degree of blurriness. Blur detection may be performed using Wavelet -based operator, which takes advantage of the capability of coefficients of the discrete wavelet transform to describe the frequency and spatial content of images. Blur detection may be performed using statistics-based operators take advantage of several image statistics as texture descriptors in order to compute a focus level. Blur detection may be performed by using discrete cosine transform (DCT) coefficients in order to compute a focus level of an image from its frequency content.

[0232] 66 Attorney Docket No. 1518-193PCT1 Continuing to refer to FIG. 7, computing device, processor, and / or module may be configured to precondition one or more training examples. For instance, and without limitation, where a machine learning model and / or process has one or more inputs and / or outputs requiring, transmitting, or receiving a certain number of bits, samples, or other units of data, one or more training examples’ elements to be used as or compared to inputs and / or outputs may be modified to have such a number of units of data. For instance, a computing device, processor, and / or module may convert a smaller number of units, such as in a low pixel count image, into a desired number of units, for instance by upsampling and interpolating. As a non-limiting example, a low pixel count image may have 100 pixels, however a desired number of pixels may be 128. Processor may interpolate the low pixel count image to convert the 100 pixels into 128 pixels. It should also be noted that one of ordinary skill in the art, upon reading this disclosure, would know the various methods to interpolate a smaller number of data units such as samples, pixels, bits, or the like to a desired number of such units. In some instances, a set of interpolation rules may be trained by sets of highly detailed inputs and / or outputs and corresponding inputs and / or outputs downsampled to smaller numbers of units, and a neural network or other machine learning model that is trained to predict interpolated pixel values using the training data. As a non-limiting example, a sample input and / or output, such as a sample picture, with sample- expanded data units (e.g., pixels added between the original pixels) may be input to a neural network or machine-learning model and output a pseudo replica sample-picture with dummy values assigned to pixels between the original pixels based on a set of interpolation rules. As a non-limiting example, in the context of an image classifier, a machine-learning model may have a set of interpolation rules trained by sets of highly detailed images and images that have been downsampled to smaller numbers of pixels, and a neural network or other machine learning model that is trained using those examples to predict interpolated pixel values in a facial picture context. As a result, an input with sample-expanded data units (the ones added between the original data units, with dummy values) may be run through a trained neural network and / or model, which may fill in values to replace the dummy values. Alternatively or additionally, processor, computing device, and / or module may utilize sample expander methods, a low-pass filter, or both. As used in this disclosure, a “low-pass filter” is a filter that passes signals with a frequency lower than a selected cutoff frequency and attenuates signals with frequencies higher than the cutoff frequency. The exact frequency response of the filter depends on the filter design.

[0233] 67 Attorney Docket No. 1518-193PCT1 Computing device, processor, and / or module may use averaging, such as luma or chroma averaging in images, to fdl in data units in between original data units.

[0234] In some embodiments, and with continued reference to FIG. 7, computing device, processor, and / or module may down-sample elements of a training example to a desired lower number of data elements. As a non-limiting example, a high pixel count image may have 256 pixels, however a desired number of pixels may be 128. Processor may down-sample the high pixel count image to convert the 256 pixels into 128 pixels. In some embodiments, processor may be configured to perform downsampling on data. Downsampling, also known as decimation, may include removing every Nth entry in a sequence of samples, all but every Nth entry, or the like, which is a process known as “compression,” and may be performed, for instance by an N-sample compressor implemented using hardware or software. Anti-aliasing and / or anti-imaging filters, and / or low-pass filters, may be used to clean up side-effects of compression.

[0235] Further referring to FIG. 7, feature selection includes narrowing and / or filtering training data to exclude features and / or elements, or training data including such elements, that are not relevant to a purpose for which a trained machine-learning model and / or algorithm is being trained, and / or collection of features and / or elements, or training data including such elements, on the basis of relevance or utility for an intended task or purpose for a trained machine-learning model and / or algorithm is being trained. Feature selection may be implemented, without limitation, using any process described in this disclosure, including without limitation using training data classifiers, exclusion of outliers, or the like.

[0236] With continued reference to FIG. 7, feature scaling may include, without limitation, normalization of data entries, which may be accomplished by dividing numerical fields by norms thereof, for instance as performed for vector normalization. Feature scaling may include absolute maximum scaling, wherein each quantitative datum is divided by the maximum absolute value of all quantitative data of a set or subset of quantitative data. Feature scaling may include min-max scaling, in which each value X has a minimum value Xminin a set or subset of values subtracted therefrom, with the result divided by the range of the values, give maximum value in the set or subset XmnxXnPW= - y Feature scaling CJ may J include mean normalization, ' which

[0237] AmaxAmin involves use of a mean value of a set and / or subset of values, Xmeanwith maximum and

[0238] 68 Attorney Docket No. 1518-193PCT1 _ minimum values: Xnew= -mean. Feature scaling may include standardization, where a

[0239] ^max~ min difference between X and Xmeanis divided by a standard deviation a of a set or subset of values: Xnew=- Scaling may be performed using a median value of a a set or subset Xmedianand / or interquartile range (IQR), which represents the difference between the 25thpercentile value and the 50thpercentile value (or closest values thereto by a rounding protocol), such as: Xnew=X Xmedian. Persons skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various alternative or additional approaches that may be used for feature scaling.

[0240] Further referring to FIG. 7, computing device, processor, and / or module may be configured to perform one or more processes of data augmentation. “Data augmentation” as used in this disclosure is addition of data to a training set using elements and / or entries already in the dataset. Data augmentation may be accomplished, without limitation, using interpolation, generation of modified copies of existing entries and / or examples, and / or one or more generative Al processes, for instance using deep neural networks and / or generative adversarial networks; generative processes may be referred to alternatively in this context as “data synthesis” and as creating “synthetic data.” Augmentation may include performing one or more transformations on data, such as geometric, color space, affine, brightness, cropping, and / or contrast transformations of images.

[0241] Still referring to FIG. 7, machine-learning module 700 may be configured to perform a lazy-learning process 720 and / or protocol, which may alternatively be referred to as a “lazy loading” or “call-when-needed” process and / or protocol, may be a process whereby machine learning is conducted upon receipt of an input to be converted to an output, by combining the input and training set to derive the algorithm to be used to produce the output on demand. For instance, an initial set of simulations may be performed to cover an initial heuristic and / or “first guess” at an output and / or relationship. As a non-limiting example, an initial heuristic may include a ranking of associations between inputs and elements of training data 704. Heuristic may include selecting some number of highest-ranking associations and / or training data 704 elements. Lazy learning may implement any suitable lazy learning algorithm, including without limitation a K-nearest neighbors algorithm, a lazy naive Bayes algorithm, or the like; persons skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various lazy- learning algorithms that may be applied to generate outputs as described in this disclosure,

[0242] 69 Attorney Docket No. 1518-193PCT1 including without limitation lazy learning applications of machine-learning algorithms as described in further detail below.

[0243] Alternatively or additionally, and with continued reference to FIG. 7, machine-learning processes as described in this disclosure may be used to generate machine-learning models 724. A “machine-learning model,” as used in this disclosure, is a data structure representing and / or instantiating a mathematical and / or algorithmic representation of a relationship between inputs and outputs, as generated using any machine-learning process including without limitation any process as described above, and stored in memory; an input is submitted to a machine-learning model 724 once created, which generates an output based on the relationship that was derived. For instance, and without limitation, a linear regression model, generated using a linear regression algorithm, may compute a linear combination of input data using coefficients derived during machine-learning processes to calculate an output datum. As a further non-limiting example, a machine-learning model 724 may be generated by creating an artificial neural network, such as a convolutional neural network comprising an input layer of nodes, one or more intermediate layers, and an output layer of nodes. Connections between nodes may be created via the process of "training" the network, in which elements from a training data 704 set are applied to the input nodes, a suitable training algorithm (such as Levenberg-Marquardt, conjugate gradient, simulated annealing, or other algorithms) is then used to adjust the connections and weights between nodes in adjacent layers of the neural network to produce the desired values at the output nodes. This process is sometimes referred to as deep learning.

[0244] Still referring to FIG. 7, machine-learning algorithms may include at least a supervised machine-learning process 728. At least a supervised machine-learning process 728, as defined herein, include algorithms that receive a training set relating a number of inputs to a number of outputs, and seek to generate one or more data structures representing and / or instantiating one or more mathematical relations relating inputs to outputs, where each of the one or more mathematical relations is optimal according to some criterion specified to the algorithm using some scoring function. For instance, a supervised learning algorithm may include ultrasonic images as described above as inputs, 3D models, such as those generated based on historical CT scan data as outputs, and a scoring function representing a desired form of relationship to be detected between inputs and outputs; scoring function may, for instance, seek to maximize the probability that a given input and / or combination of elements inputs is associated with a given

[0245] 70 Attorney Docket No. 1518-193PCT1 output to minimize the probability that a given input is not associated with a given output. Scoring function may be expressed as a risk function representing an “expected loss” of an algorithm relating inputs to outputs, where loss is computed as an error function representing a degree to which a prediction generated by the relation is incorrect when compared to a given input-output pair provided in training data 704. Persons skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various possible variations of at least a supervised machine-learning process 728 that may be used to determine relation between inputs and outputs. Supervised machine-learning processes may include classification algorithms as defined above.

[0246] With further reference to FIG. 7, training a supervised machine-learning process may include, without limitation, iteratively updating coefficients, biases, weights based on an error function, expected loss, and / or risk function. For instance, an output generated by a supervised machine-learning model using an input example in a training example may be compared to an output example from the training example; an error function may be generated based on the comparison, which may include any error function suitable for use with any machine-learning algorithm described in this disclosure, including a square of a difference between one or more sets of compared values or the like. Such an error function may be used in turn to update one or more weights, biases, coefficients, or other parameters of a machine-learning model through any suitable process including without limitation gradient descent processes, least-squares processes, and / or other processes described in this disclosure. This may be done iteratively and / or recursively to gradually tune such weights, biases, coefficients, or other parameters. Updating may be performed, in neural networks, using one or more back-propagation algorithms. Iterative and / or recursive updates to weights, biases, coefficients, or other parameters as described above may be performed until currently available training data is exhausted and / or until a convergence test is passed, where a “convergence test” is a test for a condition selected as indicating that a model and / or weights, biases, coefficients, or other parameters thereof has reached a degree of accuracy. A convergence test may, for instance, compare a difference between two or more successive errors or error function values, where differences below a threshold amount may be taken to indicate convergence. Alternatively or additionally, one or more errors and / or error function values evaluated in training iterations may be compared to a threshold.

[0247] Still referring to FIG. 7, a computing device, processor, and / or module may be configured to perform method, method step, sequence of method steps and / or algorithm

[0248] 71 Attorney Docket No. 1518-193PCT1 described in reference to this figure, in any order and with any degree of repetition. For instance, a computing device, processor, and / or module may be configured to perform a single step, sequence and / or algorithm repeatedly until a desired or commanded outcome is achieved; repetition of a step or a sequence of steps may be performed iteratively and / or recursively using outputs of previous repetitions as inputs to subsequent repetitions, aggregating inputs and / or outputs of repetitions to produce an aggregate result, reduction or decrement of one or more variables such as global variables, and / or division of a larger processing task into a set of iteratively addressed smaller processing tasks. A computing device, processor, and / or module may perform any step, sequence of steps, or algorithm in parallel, such as simultaneously and / or substantially simultaneously performing a step two or more times using two or more parallel threads, processor cores, or the like; division of tasks between parallel threads and / or processes may be performed according to any protocol suitable for division of tasks between iterations. Persons skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various ways in which steps, sequences of steps, processing tasks, and / or data may be subdivided, shared, or otherwise dealt with using iteration, recursion, and / or parallel processing.

[0249] Further referring to FIG. 7, machine learning processes may include at least an unsupervised machine-learning processes 732. An unsupervised machine-learning process, as used herein, is a process that derives inferences in datasets without regard to labels; as a result, an unsupervised machine-learning process may be free to discover any structure, relationship, and / or correlation provided in the data. Unsupervised processes 732 may not require a response variable; unsupervised processes 732 may be used to find interesting patterns and / or inferences between variables, to determine a degree of correlation between two or more variables, or the like.

[0250] Still referring to FIG. 7, machine-learning module 700 may be designed and configured to create a machine-learning model 724 using techniques for development of linear regression models. Linear regression models may include ordinary least squares regression, which aims to minimize the square of the difference between predicted outcomes and actual outcomes according to an appropriate norm for measuring such a difference (e.g., a vector-space distance norm); coefficients of the resulting linear equation may be modified to improve minimization. Linear regression models may include ridge regression methods, where the function to be minimized includes the least-squares function plus term multiplying the square of each

[0251] 72 Attorney Docket No. 1518-193PCT1 coefficient by a scalar amount to penalize large coefficients. Linear regression models may include least absolute shrinkage and selection operator (LASSO) models, in which ridge regression is combined with multiplying the least-squares term by a factor of 1 divided by double the number of samples. Linear regression models may include a multi-task lasso model wherein the norm applied in the least-squares term of the lasso model is the Frobenius norm amounting to the square root of the sum of squares of all terms. Linear regression models may include the elastic net model, a multi-task elastic net model, a least angle regression model, a LARS lasso model, an orthogonal matching pursuit model, a Bayesian regression model, a logistic regression model, a stochastic gradient descent model, a perceptron model, a passive aggressive algorithm, a robustness regression model, a Huber regression model, or any other suitable model that may occur to persons skilled in the art upon reviewing the entirety of this disclosure. Linear regression models may be generalized in an embodiment to polynomial regression models, whereby a polynomial equation (e.g., a quadratic, cubic or higher-order equation) providing a best predicted output / actual output fit is sought; similar methods to those described above may be applied to minimize error functions, as will be apparent to persons skilled in the art upon reviewing the entirety of this disclosure.

[0252] Continuing to refer to FIG. 7, machine-learning algorithms may include, without limitation, linear discriminant analysis. Machine-learning algorithm may include quadratic discriminant analysis. Machine-learning algorithms may include kernel ridge regression. Machine-learning algorithms may include support vector machines, including, without limitation, support vector classification-based regression processes. Machine-learning algorithms may include stochastic gradient descent algorithms, including classification and regression algorithms based on stochastic gradient descent. Machine-learning algorithms may include nearest neighbors algorithms. Machine-learning algorithms may include various forms of latent space regularization such as variational regularization. Machine-learning algorithms may include Gaussian processes such as Gaussian Process Regression. Machine-learning algorithms may include cross-decomposition algorithms, including partial least squares and / or canonical correlation analysis. Machine-learning algorithms may include naive Bayes methods. Machinelearning algorithms may include algorithms based on decision trees, such as decision tree classification or regression algorithms. Machine-learning algorithms may include ensemble methods such as bagging meta-estimator, forest of randomized trees, AdaBoost, gradient tree

[0253] 73 Attorney Docket No. 1518-193PCT1 boosting, and / or voting classifier methods. Machine-learning algorithms may include neural net algorithms, including convolutional neural net processes.

[0254] Still referring to FIG. 7, a machine-learning model and / or process may be deployed or instantiated by incorporation into a program, apparatus, system and / or module. For instance, and without limitation, a machine-learning model, neural network, and / or some or all parameters thereof may be stored and / or deployed in any memory or circuitry. Parameters such as coefficients, weights, and / or biases may be stored as circuit-based constants, such as arrays of wires and / or binary inputs and / or outputs set at logic “1” and “0” voltage levels in a logic circuit to represent a number according to any suitable encoding system including twos complement or the like or may be stored in any volatile and / or non-volatile memory. Similarly, mathematical operations and input and / or output of data to or from models, neural network layers, or the like may be instantiated in hardware circuitry and / or in the form of instructions in firmware, machine-code such as binary operation code instructions, assembly language, or any higher- order programming language. Any technology for hardware and / or software instantiation of memory, instructions, data structures, and / or algorithms may be used to instantiate a machinelearning process and / or model, including without limitation any combination of production and / or configuration of non-reconfigurable hardware elements, circuits, and / or modules such as without limitation ASICs, production and / or configuration of reconfigurable hardware elements, circuits, and / or modules such as without limitation FPGAs, production and / or of non- reconfigurable and / or configuration non-rewritable memory elements, circuits, and / or modules such as without limitation non-rewritable ROM, production and / or configuration of reconfigurable and / or rewritable memory elements, circuits, and / or modules such as without limitation rewritable ROM or other memory technology described in this disclosure, and / or production and / or configuration of any computing device and / or component thereof as described in this disclosure. Such deployed and / or instantiated machine-learning model and / or algorithm may receive inputs from any other process, module, and / or component described in this disclosure, and produce outputs to any other process, module, and / or component described in this disclosure.

[0255] Continuing to refer to FIG. 7, any process of training, retraining, deployment, and / or instantiation of any machine-learning model and / or algorithm may be performed and / or repeated after an initial deployment and / or instantiation to correct, refine, and / or improve the machine-

[0256] 74 Attorney Docket No. 1518-193PCT1 learning model and / or algorithm. Such retraining, deployment, and / or instantiation may be performed as a periodic or regular process, such as retraining, deployment, and / or instantiation at regular elapsed time periods, after some measure of volume such as a number of bytes or other measures of data processed, a number of uses or performances of processes described in this disclosure, or the like, and / or according to a software, firmware, or other update schedule. Alternatively or additionally, retraining, deployment, and / or instantiation may be event-based, and may be triggered, without limitation, by user inputs indicating sub-optimal or otherwise problematic performance and / or by automated field testing and / or auditing processes, which may compare outputs of machine-learning models and / or algorithms, and / or errors and / or error functions thereof, to any thresholds, convergence tests, or the like, and / or may compare outputs of processes described herein to similar thresholds, convergence tests or the like. Event-based retraining, deployment, and / or instantiation may alternatively or additionally be triggered by receipt and / or generation of one or more new training examples; a number of new training examples may be compared to a preconfigured threshold, where exceeding the preconfigured threshold may trigger retraining, deployment, and / or instantiation.

[0257] Still referring to FIG. 7, retraining and / or additional training may be performed using any process for training described above, using any currently or previously deployed version of a machine-learning model and / or algorithm as a starting point. Training data for retraining may be collected, preconditioned, sorted, classified, sanitized or otherwise processed according to any process described in this disclosure. Training data may include, without limitation, training examples including inputs and correlated outputs used, received, and / or generated from any version of any system, module, machine-learning model or algorithm, apparatus, and / or method described in this disclosure; such examples may be modified and / or labeled according to user feedback or other processes to indicate desired results, and / or may have actual or measured results from a process being modeled and / or predicted by system, module, machine-learning model or algorithm, apparatus, and / or method as “desired” results to be compared to outputs for training processes as described above.

[0258] Redeployment may be performed using any reconfiguring and / or rewriting of reconfigurable and / or rewritable circuit and / or memory elements; alternatively, redeployment may be performed by production of new hardware and / or software components, circuits, instructions, or the like, which may be added to and / or may replace existing hardware and / or

[0259] 75 Attorney Docket No. 1518-193PCT1 software components, circuits, instructions, or the like.

[0260] Further referring to FIG. 7, one or more processes or algorithms described above may be performed by at least a dedicated hardware unit 736. A “dedicated hardware unit,” for the purposes of this figure, is a hardware component, circuit, or the like, aside from a principal control circuit and / or processor performing method steps as described in this disclosure, that is specifically designated or selected to perform one or more specific tasks and / or processes described in reference to this figure, such as without limitation preconditioning and / or sanitization of training data and / or training a machine-learning algorithm and / or model. A dedicated hardware unit 736 may include, without limitation, a hardware unit that can perform iterative or massed calculations, such as matrix-based calculations to update or tune parameters, weights, coefficients, and / or biases of machine-learning models and / or neural networks, efficiently using pipelining, parallel processing, or the like; such a hardware unit may be optimized for such processes by, for instance, including dedicated circuitry for matrix and / or signal processing operations that includes, e.g., multiple arithmetic and / or logical circuit units such as multipliers and / or adders that can act simultaneously and / or in parallel or the like. Such dedicated hardware units 736 may include, without limitation, graphical processing units (GPUs), dedicated signal processing modules, FPGA or other reconfigurable hardware that has been configured to instantiate parallel processing units for one or more specific tasks, or the like, A computing device, processor, apparatus, or module may be configured to instruct one or more dedicated hardware units 736 to perform one or more operations described herein, such as evaluation of model and / or algorithm outputs, one-time or iterative updates to parameters, coefficients, weights, and / or biases, and / or any other operations such as vector and / or matrix operations as described in this disclosure.

[0261] With continued reference to FIG. 7, apparatus 100 may use user feedback to train the machine-learning models and / or classifiers described above. For example, classifier may be trained using past inputs and outputs of classifier. In some embodiments, if user feedback indicates that an output of classifier was “bad,” then that output and the corresponding input may be removed from training data used to train classifier, and / or may be replaced with a value entered by, e.g., another user that represents an ideal output given the input the classifier originally received, permitting use in retraining, and adding to training data; in either case, classifier may be retrained with modified training data as described in further detail below. In

[0262] 76 Attorney Docket No. 1518-193PCT1 some embodiments, training data of classifier may include user feedback.

[0263] With continued reference to FIG. 7, in some embodiments, an accuracy score may be calculated for classifier using user feedback. For the purposes of this disclosure, “accuracy score,” is a numerical value concerning the accuracy of a machine-learning model. For example, a plurality of user feedback scores may be averaged to determine an accuracy score. In some embodiments, a cohort accuracy score may be determined for particular cohorts of persons. For example, user feedback for users belonging to a particular cohort of persons may be averaged together to determine the cohort accuracy score for that particular cohort of persons and used as described above. Accuracy score or another score as described above may indicate a degree of retraining needed for a machine-learning model such as a classifier; apparatus 100 may perform a larger number of retraining cycles for a higher number (or lower number, depending on a numerical interpretation used), and / or may collect more training data for such retraining, perform more training cycles, apply a more stringent convergence test such as a test requiring a lower mean squared error, and / or indicate to a user and / or operator that additional training data is needed.

[0264] Referring now to FIG. 8, an exemplary embodiment of neural network 800 is illustrated. A neural network 800 also known as an artificial neural network, is a network of “nodes,” or data structures having one or more inputs, one or more outputs, and a function determining outputs based on inputs. Such nodes may be organized in a network, such as without limitation a convolutional neural network (CNN), including an input layer of nodes 804, one or more intermediate layers 808, and an output layer of nodes 812. Connections between nodes may be created via the process of "training" the network, in which elements from a training dataset are applied to the input nodes, a suitable training algorithm (such as Levenberg-Marquardt, conjugate gradient, simulated annealing, or other algorithms) is then used to adjust the connections and weights between nodes in adjacent layers of the neural network to produce the desired values at the output nodes. This process is sometimes referred to as deep learning. Connections may run solely from input nodes toward output nodes in a “feed-forward” network, or may feed outputs of one layer back to inputs of the same or a different layer in a “recurrent network.” As a further non-limiting example, a neural network may include a convolutional neural network comprising an input layer of nodes, one or more intermediate layers, and an output layer of nodes.

[0265] 77 Attorney Docket No. 1518-193PCT1 Referring now to FIG. 9, an exemplary embodiment of a node 900 of a neural network is illustrated. A node may include, without limitation, a plurality of inputs xi that may receive numerical values from inputs to a neural network containing the node and / or from other nodes.

[0266] Node may perform one or more activation functions to produce its output given one or more inputs, such as without limitation computing a binary step function comparing an input to a threshold value and outputting either a logic 1 or logic 0 output or something equivalent, a linear activation function whereby an output is directly proportional to the input, and / or a non-linear activation function, wherein the output is not proportional to the input. Non-linear activation functions may include, without limitation, a sigmoid function of the form (x) = given g X - Q ~ input x, a tanh (hyperbolic tangent) function, of the formeX+e-x, a tanh derivative function such as / (x) = tanh2(x), a rectified linear unit function such as f(x) = max (0, x), a “leaky” and / or “parametric” rectified linear unit function such as f(x) = max (ax, x) for some a, an

[0267] ( x for x 0 exponential linear units function such as f(x) = 1 ,r„ for some value of a

[0268] 1 7 Ja(ex- 1) for x < 0

[0269] (this function may be replaced and / or weighted by its own derivative in some embodiments), a exsoftmax function such as f xi) = where the inputs to an instant layer are xha swish function such as f(x) = x * sigmoid(x), a Gaussian error linear unit function such as f(x) = a(l + tanh for some values of a, b, and r, and / or a scaled exponential linear unit function such as (x) = Fundamentally, there is no limit to the nature of functions of inputs xi that may be used as activation functions. As a non-limiting and illustrative example, node may perform a weighted sum of inputs using weights w> that are multiplied by respective inputs xt. Additionally or alternatively, a bias b may be added to the weighted sum of the inputs such that an offset is added to each unit in the neural network layer that is independent of the input to the layer. The weighted sum may then be input into a function (p, which may generate one or more outputs y. Weight w, applied to an input xt may indicate whether the input is “excitatory,” indicating that it has strong influence on the one or more outputs for instance by the corresponding weight having a large numerical value, and / or a “inhibitory,” indicating it has a weak effect influence on the one more inputs y, for instance by the corresponding weight having a small numerical value. The values of weights wi may be

[0270] 78 Attorney Docket No. 1518-193PCT1 determined by training a neural network using training data, which may be performed using any suitable process as described above.

[0271] Still referring to FIG. 9, a “convolutional neural network,” as used in this disclosure, is a neural network in which at least one hidden layer is a convolutional layer that convolves inputs to that layer with a subset of inputs known as a “kernel,” along with one or more additional layers such as pooling layers, fully connected layers, and the like. CNN may include, without limitation, a deep neural network (DNN) extension, where a DNN is defined as a neural network with two or more hidden layers.

[0272] Still referring to FIG. 9, in some embodiments, a convolutional neural network may learn from images. In non-limiting examples, a convolutional neural network may perform tasks such as classifying images, detecting objects depicted in an image, segmenting an image, and / or processing an image. In some embodiments, a convolutional neural network may operate such that each node in an input layer is only connected to a region of nodes in a hidden layer. In some embodiments, the regions in aggregate may create a feature map from an input layer to the hidden layer. In some embodiments, a convolutional neural network may include a layer in which the weights and biases for all nodes are the same. In some embodiments, this may allow a convolutional neural network to detect a feature, such as an edge, across different locations in an image.

[0273] Still referring to FIG. 9, 3D Cardiac Model Generation Machine Learning Model may include a CNN, such as a 3D CNN. In some embodiments, 3D CNN, unlike a standard 2D CNN, may include utilization of one or more 3D convolutions which allow them to directly process 3D data, thereby enabling processor to generate 3D structures such as 3D data structure a heart using the 3D CNN. In a non-limiting example, 3D CNN may include one or more 3D filters (i.e., kernels) that move through ultrasonic images in three dimensions and capturing spatial relationships in x, y, and z axis. Similar to 3D convolutions, 3D CNN may further include one or more 3D pooling layers that may be used to reduce the dimensionality of ultrasonic images while preserving spatial features as described above. Additionally, or alternatively, an encoder-decoder structure may be implemented (extended to 3D) in 3D CNN, wherein the encoder-decoder structure includes an encoding path that captures the context and a decoding path that enables precise localization in a same manner as U-net as described above. Such encoder-decoder structures may also include a plurality of skip connections, allowing 3D CNN to use information

[0274] 79 Attorney Docket No. 1518-193PCT1 from multiple resolutions to improve the process of generating 3D data structure of a heart.

[0275] With continued reference to FIG. 9, in an embodiment, training a 3D cardiac model generation machine learning model may include selecting a suitable loss function to guide the training process. In a non-limiting example, a loss function that measures the difference between the predicted 3D VORs and the ground truth 3D structure e.g., CT-based 3D models may be used, such as, without limitation, mean squared error (MSE) or a custom loss function may be designed for one or more embodiments described herein. Additionally, or alternatively, optimization algorithms, such as stochastic gradient descent (SGD), may then be used to adjust the model’s parameters to minimize such loss. In a further non-limiting embodiment, instead of directly predicting 3D data structure, 3D cardiac model generation machine learning model may be trained as a regression model to predict presence indicators and / or other embedded values described herein for each voxel of plurality of voxels within a 3D grid. Additionally, CNN may be extended with additional deep learning techniques, such as recurrent neural networks (RNNs) or attention mechanism, to capture additional features and / or data relationships within input data. These extensions may further enhance the accuracy and robustness of the 3D modeling.

[0276] Referring now to FIG. 10, an exemplary embodiment of a method 1000 of determining a cardiac implant size is illustrated. One or more steps if method 1000 may be implemented, without limitation, as described with reference to other figures. One or more steps of method 1000 may be implemented, without limitation, using at least a processor.

[0277] Still referring to FIG. 10, in some embodiments, method 1000 may include a step 1005 of collecting a plurality of ultrasonic images. In some embodiments, step 1005 may be performed through use of an ultrasonic imaging device. In some embodiments, a plurality of ultrasonic images includes transesophageal echocardiogram (TEE) images. In some embodiments, a plurality of ultrasonic images includes intracardiac echo (ICE) images. In some embodiments, the plurality of ultrasonic images comprises ultrasonic images captured at the same position and different orientations with respect to a subject’s heart. In some embodiments, the plurality of ultrasonic images comprises a view of a subject’s heart at 0°, 45°, 90°, and 135° on a mid- esophageal view.

[0278] Still referring to FIG. 10, in some embodiments, method 1000 may include a step 1010 of, using a 3D cardiac model generation machine learning model trained on a training dataset comprising example ultrasonic images correlated with example 3D cardiac models, generating a

[0279] 80 Attorney Docket No. 1518-193PCT1 3D cardiac model based on the plurality of ultrasonic images. In some embodiments, the example 3D cardiac models include computed tomography (CT) scan data. In some embodiments, generating a 3D cardiac model based on the plurality of ultrasonic images includes generating a set of shape parameters representing a cardiac shape as a function of the plurality of ultrasonic images and a shape identification model trained on the training dataset; and the set of shape parameters comprises a plurality of numerical descriptors representing at least a geometric characteristic of a subject’s heart and a plurality of associated parameter ranges. In some embodiments of method 1000, step 1010 may include generating a 3D voxel occupancy representation (VOR) representing a cardiac shape as a function of plurality of ultrasonic images and 3D cardiac model generation machine learning model trained on the training dataset, generating a mesh representing the cardiac shape as a function of the 3D voxel occupancy representation, and displaying, using display, the mesh to a user.

[0280] Still referring to FIG. 10, in some embodiments, method 1000 may include a step 1015 of generating at least a cardiac measurement based on the 3D cardiac model. In some embodiments, the cardiac measurement comprises an ostial diameter of a left atrial appendage (LAA).

[0281] Still referring to FIG. 10, in some embodiments, method 1000 may include a step 1020 of determining a cardiac implant size based on the at least a cardiac measurement.

[0282] Still referring to FIG. 10, in some embodiments, method 1000 may include a step 1025 of displaying to a user the cardiac implant size. In some embodiments, a cardiac implant size comprises a left atrial appendage occlusion device size.

[0283] Still referring to FIG. 10, in some embodiments, method 1000 may further include determining a cardiac implant placement as a function of the 3D cardiac model. In some embodiments, method 1000 may further include determining a cardiac implant candidate quality as a function of the 3D cardiac model. In some embodiments, method 1000 may further include generating an example ultrasonic image of the training dataset as a function of historical computed tomography (CT) scan data.

[0284] Still referring to FIG. 10, in some embodiments, method 1000 may include receiving at least an ultrasound localization datum representing one or both of position and angle of ultrasonic imaging device, and generating, using 3D cardiac model generation machine learning model, 3D cardiac model as a function of the at least an ultrasound localization datum.

[0285] Still referring to FIG. 10, in some embodiments, method 1000 may include calculating a

[0286] 81 Attorney Docket No. 1518-193PCT1 level of uncertainty at a plurality of locations on 3D cardiac model, where the plurality of locations comprises a high uncertainty region, receiving a subsequent plurality of ultrasonic images of cardiac anatomy corresponding to the high uncertainty region of the 3D cardiac model, where the subsequent plurality of ultrasonic images is captured using ultrasonic imaging device as a function of the high uncertainty region, and generating a subsequent 3D cardiac model as a function of the subsequent plurality of ultrasonic images.

[0287] Still referring to FIG. 10, in some embodiments, method 1000 may include receiving a 3D cardiac implant model representing a cardiac implant and displaying, using a display, the 3D cardiac implant model positioned within 3D cardiac model.

[0288] Referring now to FIG. 11, a block diagram of an exemplary system 1100 for transesophageal echocardiogram guided implantation of left atrial appendage closure device is illustrated. System 1100 includes at least a transesophageal echocardiogram (TEE) system 1102. For the purposes of this disclosure, a “transesophageal echocardiography (TEE) system” is a system designed to perform transesophageal echocardiography. For the purposes of this disclosure, “transesophageal echocardiography” is a medical imaging technique in which an ultrasound transducer is inserted into the esophagus to produce images of the heart and surrounding structures. In some embodiments, TEE system 1102 may include a combination of specialized hardware and software designed to facilitate transesophageal echocardiography by positioning an ultrasound probe (e.g., ultrasound transducer) within esophagus of a patient, close to a heart of the patient. As the esophagus is proximate to the heart, a TEE system 1102 can detect ultrasound image 1104 of cardiac tissue of a patient. In some embodiments, TEE system 1102 may include a TEE probe (endoscope). The TEE probe is a long, flexible device equipped with an ultrasound transducer (ultrasound sensor 1106) at its tip. Ultrasound transducer can emit high-frequency sound waves and capture the echoes reflected from cardiac structures to produce detailed images (ultrasound image 1104). In a non-limiting example, TEE probe may be inserted into the patient’s esophagus, where the TEE probe may be connected to an ultrasound machine, which processes signals from ultrasound transducer to generate images of the heart. In some embodiments, TEE system 1102 may be communicatively connected to at least a display 1108. The display disclosed herein is further described in detail below.

[0289] With continued reference to FIG. 11, TEE system includes at least an ultrasound sensor 1106 configured to be located within an esophagus of a patient and detect at least an ultrasound

[0290] 82 Attorney Docket No. 1518-193PCT1 image 1104 as a function of cardiac tissue of the patient for the purposes of this disclosure, an “ultrasound sensor” is a device that uses ultrasonic waves. In a non-limiting example, ultrasound sensor 1106 may measure the distance to an object using ultrasonic sound waves. For the purposes of this disclosure, a “sensor” is a device that produces an output signal for the purpose of sensing a physical phenomenon. For example, and without limitation, sensor may transduce a detected phenomenon, such as without limitation, temperature, voltage, current, pressure, speed, motion, light, moisture, sound waves, and the like, into a sensed signal. Sensor may output the sensed signal. Sensor may include any computing device as described in the entirety of this disclosure and configured to convert and / or translate a plurality of signals detected into electrical signals for further analysis and / or manipulation. Electrical signals may include analog signals, digital signals, periodic or aperiodic signal, step signals, unit impulse signal, unit ramp signal, unit parabolic signal, signum function, exponential signal, rectangular signal, triangular signal, sinusoidal signal, sine function, or pulse width modulated signal. Any datum captured by sensor may include circuitry, computing devices, electronic components or a combination thereof that translates into at least an electronic signal configured to be transmitted to another electronic component. In a non-limiting embodiment, sensor may include a plurality of sensors included in a sensor suite. In one or more embodiments, and without limitation, sensor may include a plurality of sensors. Sensor may include an ultrasound sensor 1106.

[0291] With continued reference to FIG. 11, in some embodiments, ultrasound sensor 1106 may include an electrode. For the purposes of this disclosure, an “electrode” is a conductive material or element that facilitates the transmission and reception of electrical signals associated with ultrasound waves. In a non-limiting example, electrode may detect and record electrical activity; for instance, but not limited to, the heart's electrical signals. For example, and without limitation, electrode may generate ultrasonic sound waves, from which ultrasound sensor 1106 receives the ultrasonic waves and transmit ultrasound image 1104 related to the ultrasonic waves to processor 1110. In some embodiments, ultrasound sensor 1106 may include a transducer. For the purposes of this disclosure, a “transducer” is a component of an ultrasound sensor that converts one form of energy into another. In a non-limiting example, transducer may operate on a principle of piezoelectricity, where piezoelectric material can convert electrical energy into mechanical vibration (i.e., ultrasonic waves) and vice versa. In some embodiments, ultrasound sensor 1106 may include a transceiver. For the purposes of this disclosure, a “transceiver” is a combined unit

[0292] 83 Attorney Docket No. 1518-193PCT1 of a transmitter and a receiver. In a non-limiting example, transceiver may transmit ultrasonic waves and receive echoes.

[0293] With continued reference to FIG. 11, for the purposes of this disclosure, “ultrasound image” is a visual representation generated by reflection of high-frequency sound waves off internal body structures. In a non-limiting example, ultrasound image 1104 may include visual representation of a heart examined through esophagus. As a non-limiting example, ultrasound image 1104 may include distance between sensor and surrounding tissue or organs. In some cases, ultrasound sensor 1106 may detect ultrasound image 1104 in a plurality of angles. In a non-limiting example, ultrasound image 1104 may include a plurality of distances between sensor and a heart in different angles. For example, and without limitation, when ultrasound sensor 1106 moves around within an esophagus, ultrasound sensor 1106 receives a plurality of distances between ultrasound sensor 1106 and organ and generate ultrasound image 1104 using the plurality of distances. In a non-limiting example, ultrasound image 1104 may include an image of a heart before LAAC device implantation, during the LAAC device implantation and after the LAAC device implantation. For example, and without limitation, ultrasound image 1104 may include an image of a left atrium appendage of a heart. For example, and without limitation, ultrasound image 1104 may include an image of a heart and a catheter entering into a left atrium appendage of the heart. For example, and without limitation, ultrasound image 1104 may include an image of left atrium appendage of a heart with a left atrium appendage closure device implanted within the left atrium appendage. As another non-limiting example, ultrasound sensor 1106 may include signal strength or amplitude of ultrasonic signal emitted and received by ultrasound sensor 1106, images within an organ, or the like. As another non-limiting example, ultrasound sensor 1106 may include ambient temperature, humidity, atmospheric pressure, or the like. In some embodiments, ultrasound sensor 1106 may be stored in a database. In some embodiments, ultrasound image 1104 may be retrieved from database. In some embodiments, user may manually input ultrasound image 1104. In some embodiments, ultrasound image 1104 may be received from remote device. As a non-limiting example, processor 1110 may receive ultrasound sensor 1106 from a computing device or processor incorporated with ultrasound sensor 1106 or TEE system 1102.

[0294] With continued reference to FIG. 11, system 1100 includes a computing device 1112. Computing device 1112 includes a processor 1110 communicatively connected to a memory

[0295] 84 Attorney Docket No. 1518-193PCT1 1114. As used in this disclosure, “communicatively connected” means connected by way of a connection, attachment or linkage between two or more relata which allows for reception and / or transmittance of information therebetween. For example, and without limitation, this connection may be wired or wireless, direct or indirect, and between two or more components, circuits, devices, systems, and the like, which allows for reception and / or transmittance of data and / or signal(s) therebetween. Data and / or signals therebetween may include, without limitation, electrical, electromagnetic, magnetic, video, audio, radio and microwave data and / or signals, combinations thereof, and the like, among others. A communicative connection may be achieved, for example and without limitation, through wired or wireless electronic, digital or analog, communication, either directly or by way of one or more intervening devices or components. Further, communicative connection may include electrically coupling or connecting at least an output of one device, component, or circuit to at least an input of another device, component, or circuit. For example, and without limitation, via a bus or other facility for intercommunication between elements of a computing device 1112. Communicative connecting may also include indirect connections via, for example without limitation, wireless connection, radio communication, low power wide area network, optical communication, magnetic, capacitive, or optical coupling, and the like. In some instances, the terminology “communicatively coupled” may be used in place of communicatively connected in this disclosure.

[0296] With continued reference to FIG. 11, in some embodiments, computing device 1112 may include any computing device as described in this disclosure, including without limitation a microcontroller, microprocessor, digital signal processor (DSP) and / or system on a chip (SoC) as described in this disclosure. Computing device 1112 may include, be included in, and / or communicate with a mobile device such as a mobile telephone or smartphone. Computing device 1112 may include a single computing device operating independently, or may include two or more computing devices operating in concert, in parallel, sequentially or the like; two or more computing devices may be included together in a single computing device or in two or more computing devices. Computing device 1112 may interface or communicate with one or more additional devices as described below in further detail via a network interface device. Network interface device may be utilized for connecting computing device 1112 to one or more of a variety of networks, and one or more devices. Examples of a network interface device include,

[0297] 85 Attorney Docket No. 1518-193PCT1 but are not limited to, a network interface card (e.g., a mobile network interface card, a LAN card), a modem, and any combination thereof. Examples of a network include, but are not limited to, a wide area network (e.g., the Internet, an enterprise network), a local area network (e.g., a network associated with an office, a building, a campus or other relatively small geographic space), a telephone network, a data network associated with a telephone / voice provider (e.g., a mobile communications provider data and / or voice network), a direct connection between two computing devices, and any combinations thereof. A network may employ a wired and / or a wireless mode of communication. In general, any network topology may be used. Information (e.g., data, software etc.) may be communicated to and / or from a computer and / or a computing device 1112. Computing device 1112 may include but is not limited to, for example, a computing device or cluster of computing devices in a first location and a second computing device or cluster of computing devices in a second location. Computing device 1112 may include one or more computing devices dedicated to data storage, security, distribution of traffic for load balancing, and the like. Computing device 1112 may distribute one or more computing tasks as described below across a plurality of computing devices of computing device, which may operate in parallel, in series, redundantly, or in any other manner used for distribution of tasks or memory between computing devices. Computing device 1112 may be implemented, as a nonlimiting example, using a “shared nothing” architecture.

[0298] With continued reference to FIG. 11, in some embodiments, computing device 1112 may be designed and / or configured to perform any method, method step, or sequence of method steps in any embodiment described in this disclosure, in any order and with any degree of repetition. For instance, computing device 1112 may be configured to perform a single step or sequence repeatedly until a desired or commanded outcome is achieved; repetition of a step or a sequence of steps may be performed iteratively and / or recursively using outputs of previous repetitions as inputs to subsequent repetitions, aggregating inputs and / or outputs of repetitions to produce an aggregate result, reduction or decrement of one or more variables such as global variables, and / or division of a larger processing task into a set of iteratively addressed smaller processing tasks. Computing device 1112 may perform any step or sequence of steps as described in this disclosure in parallel, such as simultaneously and / or substantially simultaneously performing a step two or more times using two or more parallel threads, processor cores, or the like; division of tasks between parallel threads and / or processes may be performed according to any protocol

[0299] 86 Attorney Docket No. 1518-193PCT1 suitable for division of tasks between iterations. Persons skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various ways in which steps, sequences of steps, processing tasks, and / or data may be subdivided, shared, or otherwise dealt with using iteration, recursion, and / or parallel processing.

[0300] With continued reference to FIG. 11, memory 1114 contains instructions configuring processor 1110 to receive at least an ultrasound image 1104. In some embodiments, processor 1110 may receive ultrasound image 1104 from TEE system 1102. In some embodiments, processor 1110 may receive ultrasound image 1104 from a left atrial appendage closure (LAAC) database 1116. In some embodiments, system 1100 may include a LAAC database 1116. As used in this disclosure, “left atrial appendage closure database” is a data structure configured to store data associated with a left atrial appendage closure. In one or more embodiments, LAAC database 1116 may include inputted or calculated information and datum related to left atrial appendage closure. In some embodiments, a datum history may be stored in LAAC database 1116. As a non-limiting example, the datum history may include real-time and / or previous inputted data related to left atrial appendage closure. As a non-limiting example, LAAC database 1116 may include instructions from a user, who may be an expert user, a past user in embodiments disclosed herein, or the like, where the instructions may include examples of the data related to left atrial appendage closure.

[0301] With continued reference to FIG. 11, in some embodiments, processor 1110 may be communicatively connected with LAAC database 1116. For example, and without limitation, in some cases, LAAC database 1116 may be local to processor 1110. In another example, and without limitation, LAAC database 1116 may be remote to processor 1110 and communicative with processor 1110 by way of one or more networks. The network may include, but is not limited to, a cloud network, a mesh network, and the like. By way of example, a “cloud-based” system can refer to a system which includes software and / or data which is stored, managed, and / or processed on a network of remote servers hosted in the “cloud,” e.g., via the Internet, rather than on local severs or personal computers. A “mesh network” as used in this disclosure is a local network topology in which the infrastructure processor 1110 connect directly, dynamically, and non-hierarchically to as many other computing devices as possible. A “network topology” as used in this disclosure is an arrangement of elements of a communication network. The network may use an immutable sequential listing to securely store LAAC database 1116. An

[0302] 87 Attorney Docket No. 1518-193PCT1 “immutable sequential listing,” as used in this disclosure, is a data structure that places data entries in a fixed sequential arrangement, such as a temporal sequence of entries and / or blocks thereof, where the sequential arrangement, once established, cannot be altered or reordered. An immutable sequential listing may be, include and / or implement an immutable ledger, where data entries that have been posted to the immutable sequential listing cannot be altered.

[0303] With continued reference to FIG. 11, in some embodiments, LAAC database 1116 may be implemented, without limitation, as a relational database, a key-value retrieval database such as a NOSQL database, or any other format or structure for use as a database that a person skilled in the art would recognize as suitable upon review of the entirety of this disclosure. Database may alternatively or additionally be implemented using a distributed data storage protocol and / or data structure, such as a distributed hash table or the like. Database may include a plurality of data entries and / or records as described above. Data entries in a database may be flagged with or linked to one or more additional elements of information, which may be reflected in data entry cells and / or in linked tables such as tables related by one or more indices in a relational database. Persons skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various ways in which data entries in a database may store, retrieve, organize, and / or reflect data and / or records as used herein, as well as categories and / or populations of data consistently with this disclosure.

[0304] With continued reference to FIG. 11, in some embodiments, processor 1110 may receive ultrasound image 1104 from a user device 1118. For the purposes of this disclosure, a “user device” is any device a user use to input data. For the purposes of this disclosure, a “user” is an individual or entity that uses an apparatus. As a non-limiting example, a user may include a surgeon, doctor, medical professional, and the like. As a non-limiting example, user device 1118 may include a laptop, desktop, tablet, mobile phone, smart phone, smart watch, kiosk, screen, smart headset, or things of the like. In some embodiments, user device 1118 may include an interface configured to receive inputs from user. In some embodiments, user may manually input any data into computing device 1112 using user device 1118. In some embodiments, user may have a capability to process, store or transmit any information independently.

[0305] With continued reference to FIG. 11, in some embodiments, receiving at least an ultrasound image 1104 may include extracting at least a TEE angle datum 1120 from at least an ultrasound image 1104 using an optical character recognition. For the purposes of this disclosure,

[0306] 88 Attorney Docket No. 1518-193PCT1 a “transesophageal echocardiogram angle datum” is a data element indicating a value that represents the angular orientation of an ultrasound sensor In some embodiments, TEE angle datum 1120 may be angular orientation of an ultrasound sensor 1106 relative to a reference axis or plane (e.g., the anatomical position of a heart). As a non-limiting example, TEE angle datum 1120 may include TEE probe's imaging angle, such as 0°, 45°, 90°, or 135°, which may determine the plane of the ultrasound slice captured during imaging.

[0307] With continued reference to FIG. 11, in some embodiments, processor 1110 may analyze ultrasound image 1104 to find TEE angle datum 1120 using optical character recognition (OCR) 1122. In some embodiments, ultrasound image 1104 may include a plurality of words related to position and orientation of ultrasound sensor 1106 within esophagus. For the purposes of this disclosure, “optical character recognition” is a technology that enables the recognition and conversion of printed or written text into machine-encoded text. In some cases, the at least a processor 1110 may be configured to recognize a keyword using the OCR 1122 to find the TEE angle datum 1120. As used in this disclosure, a “keyword” is an element of word or syntax used to identify and / or match elements to each other. In some cases, the at least a processor 1110 may transcribe much or even substantially all ultrasound images 1104.

[0308] With continued reference to FIG. 11, in some embodiments, optical character recognition 1122 or optical character reader (OCR) may include automatic conversion of images of written (e g., typed, handwritten or printed text) into machine-encoded text. In some cases, recognition of a keyword from ultrasound image 1104 may include one or more processes, including without limitation optical character recognition (OCR) 1122, optical word recognition, intelligent character recognition, intelligent word recognition, and the like. In some cases, OCR 1122 may recognize written text, one glyph or character at a time. In some cases, optical word recognition may recognize written text, one word at a time, for example, for languages that use a space as a word divider. In some cases, intelligent character recognition (ICR) may recognize written text one glyph or character at a time, for instance by employing machine-learning processes. In some cases, intelligent word recognition (IWR) may recognize written text, one word at a time, for instance by employing machine-learning processes.

[0309] With continued reference to FIG. 11, in some cases, OCR 1122 may be an "offline" process, which analyses a static document or image frame. In some cases, handwriting movement analysis can be used as input to handwriting recognition. For example, instead of

[0310] 89 Attorney Docket No. 1518-193PCT1 merely using shapes of glyphs and words, this technique may capture motions, such as the order in which segments are drawn, the direction, and the pattern of putting the pen down and lifting it. This additional information may make handwriting recognition more accurate. In some cases, this technology may be referred to as “online” character recognition, dynamic character recognition, real-time character recognition, and intelligent character recognition.

[0311] With continued reference to FIG. 11, in some cases, OCR processes may employ preprocessing of ultrasound image 1104. Pre-processing process may include without limitation deskew, de-speckle, binarization, line removal, layout analysis or “zoning,” line and word detection, script recognition, character isolation or “segmentation,” and normalization. In some cases, a de-skew process may include applying a transform (e.g., homography or affine transform) to the ultrasound image 1104 to align text. In some cases, a de-speckle process may include removing positive and negative spots and / or smoothing edges. In some cases, a binarization process may include converting an image from color or greyscale to black-and-white (i.e., a binary image). Binarization may be performed as a simple way of separating text (or any other desired image component) from a background of image component. In some cases, binarization may be required for example if an employed OCR algorithm only works on binary images. In some cases, a line removal process may include removal of non-glyph or noncharacter imagery (e.g., boxes and lines). In some cases, a layout analysis or “zoning” process may identify columns, paragraphs, captions, and the like as distinct blocks. In some cases, a line and word detection process may establish a baseline for word and character shapes and separate words, if necessary. In some cases, a script recognition process may, for example in multilingual documents, identify script allowing an appropriate OCR algorithm to be selected. In some cases, a character isolation or “segmentation” process may separate signal characters, for example character-based OCR algorithms. In some cases, a normalization process may normalize aspect ratio and / or scale of image component.

[0312] With continued reference to FIG. 11, in some embodiments an OCR process may include an OCR algorithm. Exemplary OCR algorithms include matrix matching process and / or feature extraction processes. Matrix matching may involve comparing an image to a stored glyph on a pixel-by-pixel basis. In some case, matrix matching may also be known as “pattern matching,” “pattern recognition,” and / or “image correlation.” Matrix matching may rely on an input glyph being correctly isolated from the rest of the image component. Matrix matching may also rely on

[0313] 90 Attorney Docket No. 1518-193PCT1 a stored glyph being in a similar font and at a same scale as input glyph. Matrix matching may work best with typewritten text.

[0314] With continued reference to FIG. 11, in some embodiments, an OCR process may include a feature extraction process. In some cases, feature extraction may decompose a glyph into a feature. Exemplary non-limiting features may include corners, edges, lines, closed loops, line direction, line intersections, and the like. In some cases, feature extraction may reduce dimensionality of representation and may make the recognition process computationally more efficient. In some cases, extracted feature may be compared with an abstract vector-like representation of a character, which might reduce to one or more glyph prototypes. General techniques of feature detection in computer vision are applicable to this type of OCR 1122. In some embodiments, machine-learning processes like nearest neighbor classifiers (e.g., k-nearest neighbors algorithm) may be used to compare image features with stored glyph features and choose a nearest match. OCR 1122 may employ any machine-learning process described in this disclosure, for example machine-learning processes described with reference to FIG. 7. Exemplary non-limiting OCR software may include Cuneiform and Tesseract. Cuneiform may include a multi-language, open-source optical character recognition system originally developed by Cognitive Technologies of Moscow, Russia. Tesseract may include free OCR software originally developed by Hewlett-Packard of Palo Alto, California, United States.

[0315] With continued reference to FIG. 11, in some cases, OCR 1122 may employ a two-pass approach to character recognition. A first pass may try to recognize a character. Each character that is satisfactory may be passed to an adaptive classifier as training data. The adaptive classifier then may get a chance to recognize characters more accurately as it further analyzes ultrasound image 1104. Since the adaptive classifier may have learned something useful a little too late to recognize characters on the first pass, a second pass may be run over the ultrasound image 1104. Second pass may include adaptive recognition and use characters recognized with high confidence on the first pass to recognize better remaining characters on the second pass. In some cases, two-pass approach may be advantageous for unusual fonts or low-quality image components where visual verbal content may be distorted. Another exemplary OCR software tool may include OCRopus. OCRopus development is led by German Research Centre for Artificial Intelligence in Kaiserslautern, Germany. In some cases, OCR software may employ neural networks.

[0316] 91 Attorney Docket No. 1518-193PCT1 With continued reference to FIG. 11 , in some cases, OCR 1122 may include postprocessing. For example, OCR accuracy may be increased, in some cases, if output is constrained by a lexicon. A lexicon may include a list or set of words that are allowed to occur in a document. In some cases, a lexicon may include, for instance, all the words in the English language, or a more technical lexicon for a specific field. In some cases, an output stream may be a plain text stream or file of characters. In some cases, an OCR process may preserve an original layout of visual verbal content. In some cases, near-neighbor analysis can make use of cooccurrence frequencies to correct errors, by noting that certain words are often seen together. For example, “Washington, D.C.” is generally far more common in English than “Washington DOC.” In some cases, an OCR process may make us of a priori knowledge of grammar for a language being recognized. For example, grammar rules may be used to help determine if a word is likely to be a verb or a noun. Distance conceptualization may be employed for recognition and classification. For example, a Levenshtein distance algorithm may be used in OCR postprocessing to further optimize results.

[0317] With continued reference to FIG. 11, in some embodiments, processor 1110 may generate an image inquiry datum 1124 as a function of the at least a TEE angle datum 1120 and device instruction for use (IFU) data 1126 retrieved from an LAAC database 1116. For the purposes of this disclosure, an “image inquiry datum” is a data element that inquires additional ultrasound images. In some embodiments, image inquiry datum 1124 may include a control signal that is transmitted to TEE system 1102 to obtain necessary additional ultrasound images 1104. In a non-limiting example, if processor 1110 determines that ultrasound images in specific views or specific angles and / or a number of ultrasound images are missing based on device IFU data 1126, processor 1110 may request additional ultrasound images 1104 from TEE system 1102. In some embodiments, image inquiry datum 1124 may include a notification for a user. In a non-limiting example, if processor 1110 determines that ultrasound images in specific views or specific angles and / or a number of ultrasound images are missing based on device IFU data 1126, processor 1110 may generate a notification and may transmit to a user to operate TEE system 1102 to generate additional ultrasound images 1104 more and / or in the specific views or specific angles. In a non-limiting example, image inquiry datum 1124 may be configured to query additional ultrasound images 1104 to a user of at least a TEE system 1102 through at least a display 1108.

[0318] 92 Attorney Docket No. 1518-193PCT1 With continued reference to FIG. 11 , for the purposes of this disclosure, “device Instructions for Use (IFU) data” is data related to procedural guidelines or specifications associated with a left atrial appendage closure device. As a non-limiting example, device IFU data 1126 may include usage instructions, imaging requirements, and the like. For example, and without limitation, device IFU data 1126 may include specific views or angles of ultrasound images 1104 necessary for a placement of LAAC device to LAA. For example, and without limitation, device IFU data 1126 may include a number of ultrasound images 1104 necessary for a placement of LAAC device to LAA. In some embodiments, device IFU data 1126 may be stored in a LAAC database 1116. In some embodiments, processor 1110 may retrieve device IFU data 1126 from LAAC database 1116 as a function of device datum 1128. As a non-limiting example, processor 1110 may retrieve device IFU data 1126 that is related to a LAAC device selected specifically for a patient. For the purposes of this disclosure, a “left atrial atrium closure device” is a medical implant designed to occlude a left atrial appendage of a heart. In some embodiments, LAAC device may reduce the risk of thromboembolism in patients with conditions such as atrial fibrillation (AF). By sealing the LAA, LAAC device 1130 may prevent the formation or migration of blood clots that could lead to stroke, particularly in patients who are contraindicated for long-term anti coagulation therapy. As a non-limiting example, LAAC device 1130 may include Watchman device, Amplatzer Amulet, and the like.

[0319] With continued reference to FIG. 11, in some embodiments, receiving at least an ultrasound image 1104 may include generating view training data 1132, wherein the view training data 1132 may include exemplary ultrasound images correlated to exemplary view labels, training a view classifier 1134 using the view training data 1132, classifying the at least an ultrasound image 1104 to at least a view label 1136 using the trained view classifier 1134 and generating an image inquiry datum 1124 as a function of the view label 1136 and device IFU data 1126. For the purposes of this disclosure, a “view label” is an identifier associated with an imaging perspective or orientation of a heart captured during a transesophageal echocardiography. As a non-limiting example, view label 1136 may include mid-esophageal four-chamber view, mid-esophageal bicaval view, mid-esophageal long-axis view, mid- esophageal left atrial appendage view, transgastric short-axis view, transgastric two-chamber view, aortic valve short-axis view, and the like. In some embodiments, view label 1136 may be retrieved from LAAC database 1116. In some embodiments, user may manually input view label

[0320] 93 Attorney Docket No. 1518-193PCT1 1136 of ultrasound images 1104.

[0321] With continued reference to FIG. 11, in some embodiments, view training data 1132 may be stored in LAAC database 1116. For the purposes of this disclosure, “view training data” is data containing correlations that a machine-learning process may use to model relationships between ultrasound images and view labels. In some embodiments, view training data 1132 may be received from one or more users, LAAC database 1116, external computing devices, and / or previous iterations of processing. As a non-limiting example, view training data 1132 may include instructions from a user, who may be an expert user, a past user in embodiments disclosed herein, or the like, which may be stored in memory and / or stored in LAAC database 1116, where the instructions may include labeling of training examples. In some embodiments, view training data 1132 may be updated iteratively on a feedback loop. As a non-limiting example, processor 1110 may update view training data 1132 iteratively through a feedback loop as a function of ultrasound image 1104, TEE angle datum 1120, view label 1136, device IFU data 1126, or the like. In some embodiments, processor 1110 may be configured to generate a view classifier 1134. For the purposes of this disclosure, a “view classifier” is a machinelearning model as defined below, such as a data structure representing and / or using a mathematical model, neural net, or program generated by a machine learning algorithm known as a “classification algorithm,” as described in further detail below, that sorts ultrasound images into view labels. In a non-limiting example, generating view classifier 1134 may include training, retraining, or fine-tuning view classifier 1134 using view training data 1132 or updated view training data 1132. In some embodiments, processor 1110 may be configured to determine view label 1136 using view classifier 1134 (i.e., trained or updated view classifier 1134). In some embodiments, generating training data and training machine-learning models may be simultaneous.

[0322] With continued reference to FIG. 11, memory 1114 contains instructions configuring processor 1110 to generate at least a 3D cardiac model 1137 representative of a patient’s heart as a function of at least an ultrasound image 1104. For the purposes of this disclosure, a “three- dimensional cardiac model” is a three-dimensional representation of a patient’s heart and / or surrounding structures. In some embodiments, 3D cardiac model may include peripheral vasculature. “Peripheral vasculature,” for the purposes of this disclosure, is a structure or structures of blood vessels surrounding or in the immediate vicinity of the heart. In a non-

[0323] 94 Attorney Docket No. 1518-193PCT1 limiting example, 3D cardiac model 1137 may include a 3D voxel occupancy representation (VOR). As used in this disclosure, a “3D voxel occupancy representation (VOR)” of an anatomical object is a 3D digital representation of a spatial structure of the anatomical object, wherein the representation is composed of a plurality of discrete volumetric elements known as voxels. A “voxel,” for the purpose of this disclosure, is a 3D equivalent of a pixel in 2D imaging. While a pixel represents a point in a 2D image and may include properties such as color and / or brightness, a voxel may represent a volume in a 3D space and may include additional properties such density / occupancy as described below. In an embodiment, each voxel of plurality of voxels within 3D VOR may represent a specific portion of heart. In some cases, voxel may be a smallest distinguishable box-shaped part (i.e., Ipx Ipx- Ipx) of a three-dimensional image. In some cases, each voxel of plurality of voxels within VOR may be represented as a cube or rectangular prism (although other shapes may be used in specialized applications). Each voxel may include a size that determines a resolution of the 3D image or model. In an embodiment, smaller voxels may provide higher resolution; however, it may require more computational resources (e.g., RAM) for processor 1110 to process.

[0324] In an embodiment, and still referring to FIG. 11, each voxel of plurality of voxels within VOR may include one or more embedded values. As used herein, “embedded values” refers to specific numerical or categorical data associated with each voxel. In some cases, embedded values may represent various attributes or characteristics of the corresponding portion of heart that voxel represents. In a non-limiting example, embedded values may include density values, intensity values, texture information, or any other quantitative measures that provide insights into the underlying cardiac tissue. Such embedded values may be derived from set of ultrasonic images or other imaging modalities used to generate 3D cardiac model 1137. In some cases, embedded values may be utilized, by processor 1110, to differentiate between different types of cardiac tissues, such as myocardial tissue, blood vessels, or chambers. Embedded values may also facilitate the visualization of dynamic cardiac functions, for example, and without limitation, blood flow or heart beating by encoding temporal information such as timestamps within plurality of voxels.

[0325] In some cases, and still reference to FIG. 11, one or more embedded values, such as, without limitations, occupancy, or density, may be derived from ultrasound images 1104 described herein by processor 1110. In a non-limiting example, determining occupancy status of

[0326] 95 Attorney Docket No. 1518-193PCT1 each voxel of plurality of voxels may include converting set of ultrasonic images 1104 to a set of binary images and determining occupancy status of each voxel as a function of the structure of interest’s binary value. In some cases, occupancy status may include a value representing the likelihood of occupancy of the corresponding heart tissue. In another non-limiting example, density may be calculated, by processor 1110, for each voxel as a function of the echogenicity of one or more pixels on a given ultrasound image 1104, wherein, the brightness of the given ultrasonic image may be analyzed since different tissues reflect ultrasound waves differently.

[0327] With continued reference to FIG. 11, generating 3D cardiac model 1137 of heart may include generating a 3D array. In some cases, processor 1110 may divide 3D space into a grid of plurality of voxels, each with specific x, y, and z coordinates as embedded values. Each element of 3D array may correspond to a voxel. In some cases, 3D array may allow for easy access and manipulation of plurality of voxels, enabling various analyses, visualizations, and transformations either described or not described herein. In a non-limiting example, embedded values may include a density of the tissue at a specific location of a patient’s body derived from one or more ultrasonic images of ultrasound images 1104.

[0328] Additionally, or alternatively, and still referring to FIG. 11, 3D cardiac model 1137 of heart may include a 3D grid embedded values described herein of plurality of voxels (e.g., tissue density, blood flow velocity, echogenicity or acoustic properties, and any other biophysical properties). As used in this disclosure, a “3D grid” refers to a 3D data structure that divides a given volume (e.g., volume of a heart) into a plurality of discrete units called cells (i.e., volume elements). In an embodiment, each cell within 3D grid may be associated with a distinct voxel.

[0329] In yet another embodiment, and still referring to FIG. 11, cells may be continuous, meaning that one or more cells may represent one or more continuous regions of space rather than discreate, separate units. In a non-limiting example, instead of being uniform, mapped presence indicator and / or other embedded values may vary continuously across different cells or cell’s volume. In such embodiment, processor 1110 may use interpolation to estimate other (unknown) embedded values within a range based on existing values such as known embedded values at specific points, thereby allowing for smooth transitions between cells. Exemplary interpolation methods may include, without limitation, linear interpolation, cubic interpolation, and / or the like. For example, and without limitation, if the corners of a cell have known values interpolation can be used to estimate the values at any point within the cell based on those comer

[0330] 96 Attorney Docket No. 1518-193PCT1 values.

[0331] In a non-limiting example, and still referring to FIG. 11, 3D cardiac model 1137 of heart may include a 3D grid having a plurality of cells e.g., voxels, wherein each cell may contain a continuous range of values representing tissue density, blood flow velocity, or other properties (i.e., embedded values). Processor 1110 may be configured to apply trilinear or tricubic interpolation to estimate tissue density within each cell based on presence indicator or other known values at the cell’s boundaries, since tissue densities change gradually; Such 3D grid may provide a smooth, continuous representation of heat’s internal structures, allowing for more nuanced analysis and visualization as described below. In a further embodiment, 3D grid with continuous cells may be additionally used in fluid dynamics simulations.

[0332] With continued reference to FIG. 11, in some case, embedded values may be mapped to 3D grid as a function of array masking, wherein specific array or grid may be selected to modify based on one or more pre-defined criteria. In a non-limiting example, processor 1110 may generate a mask e.g., a binary array that defines which voxels or cells are affected. Mask may be used to select or modify specific voxels or cells based on certain attributes; for instance, and without limitation, processor 1110 may use mask to isolate the left atrium (LA) within the heart focusing the analysis on that specific region. Such mask may include a criteria defined by specific density thresholds that distinguish the LA’s tissue (i.e., voxels representing LA in 3D grid) from surrounding structures (i.e., neighboring voxels). In some cases, such mask may further include a binary mask, wherein each voxel in the 3D gird may be assigned a first presence indicator such as 1 if the voxel meets the criteria for the LA and a second presence indicator such as 0 if it does not. In some embodiments, mask may be directly applied to 3D grid, selecting, or modifying voxels or cells, thereby enabling processor 1110 to highlight, exclude, or otherwise manipulate specific parts of heart within 3D grid. Processor 1110 may then perform an element-wise multiplication between 3D grid and the mask. Continuing from the previous nonlimiting example, voxels corresponding to the LA (wherein the mask value is 1) may retain their original values, while other voxels (where the mask value is 0) may be set to 0 or other specific value (i.e., excluded or masked out).

[0333] With continued reference to FIG. 11, in some embodiments, 3D grid may include one or more cardiac feature datums 1138 extracted from ultrasound image 1104 of heart. As used in this disclosure, “cardiac feature datums” are specific characteristics or attributes related to a heart. In

[0334] 97 Attorney Docket No. 1518-193PCT1 some embodiments, cardiac feature datum 1 138 may include spatial arrangement, shape, size, or texture of a heart. In some cases, cardiac feature datums 1138 may include one or more embedded values described herein and their combinations thereof. In a non-limiting example, cardiac feature datum 1138 may be represented numerically as a vector, a metric or other mathematical constructs that capture specific spatial characteristics. In some cases, cardiac feature datums 1138 may also be visualized as contours, surfaces, or other geometric representations. In an embodiment, cardiac feature datums 1138 may be extracted using edge detection, texture analysis, or other image processing techniques (e.g., cleaning and enhancing images, image segmentation, and / or the like). In another embodiment, one or more machine learning models, such as convolutional neural networks (CNNs) as described in further detail below, may be used to extract complex cardiac feature datums 1138. The cardiac feature datum 1138 is further described below.

[0335] Still referring to FIG. 11, as used in this disclosure, a “vector” is a data structure that represents one or more a quantitative values and / or measures of one or more cardiac feature datums 1138. A vector may be represented as an n-tuple of values, where n is one or more values, as described in further detail below; a vector may alternatively or additionally be represented as an element of a vector space, defined as a set of mathematical objects that can be added together under an operation of addition following properties of associativity, commutativity, existence of an identity element, and existence of an inverse element for each vector, and can be multiplied by scalar values under an operation of scalar multiplication compatible with field multiplication, and that has an identity element is distributive with respect to vector addition, and is distributive with respect to field addition. Each value of n-tuple of values may represent a measurement or other quantitative value associated with a given category of data, or attribute, examples of which are provided in further detail below; a vector may be represented, without limitation, in n-dimensional space using an axis per category of value represented in n-tuple of values, such that a vector has a geometric direction characterizing the relative quantities of attributes in the n-tuple as compared to each other. Two vectors may be considered equivalent where their directions, and / or the relative quantities of values within each vector as compared to each other, are the same; thus, as a non-limiting example, a vector represented as [5, 10, 15] may be treated as equivalent, for purposes of this disclosure, as a vector represented as [1, 2, 3], Vectors may be more similar where their directions are more

[0336] 98 Attorney Docket No. 1518-193PCT1 similar, and more different where their directions are more divergent, for instance as measured using cosine similarity as computed using a dot product of two vectors; however, vector similarity may alternatively or additionally be determined using averages of similarities between like attributes, or any other measure of similarity suitable for any n-tuple of values, or aggregation of numerical similarity measures for the purposes of loss functions as described in further detail below. Any vectors as described herein may be scaled, such that each vector represents each attribute along an equivalent scale of values. Each vector may be “normalized,” or divided by a “length” attribute, such as a length attribute I as derived using a Pythagorean norm: I — j^t=oat2, where at is attribute number i of the vector. Scaling and / or normalization may function to make vector comparison independent of absolute quantities of attributes, while preserving any dependency on similarity of attributes.

[0337] Still referring to FIG. 11, in a non-limiting example, one or more cardiac feature datums 1138 may include one or more shape features (i.e., characteristics related to the shape of specific cardiac structures), such as curvature, surface area, volume, and / or the like. In another nonlimiting example, one or more cardiac feature datums 1138 may include one or more texture features (i.e., characteristics related to the texture or pattern within cardiac tissues, as seen ultrasound image 1104), such as gray-level co-occurrence matrix (GLCM) features representing the texture of heart muscle tissue. In another non-limiting example, one or more cardiac feature datums 1138 may include one or more orientation features (i.e., characteristics related to the orientation or alignment of cardiac structures), such as the angle or alignment of the septum within the heart. In a further non-limiting example, one or more cardiac feature datums 1138 may include one or more edge and boundary features (i.e., Characteristics related to the edges or boundaries between different cardiac structures or tissues), such as edge detection features highlighting the boundary between the myocardium and the cardiac chambers. As an ordinary person skilled in the art, upon reviewing the entirety of this disclosure, will be aware of various cardiac feature datums extracted from ultrasound image 1104 in consistent with this disclosure.

[0338] With continued reference to FIG. 11 , in some embodiments, system 1100 may include a computer vision model configured to generate 3D cardiac model 1137 of heart. A “computer vision model,” for the purpose of this disclosure, is a computation model designed to interpret and make determinations based on visual data. In an embodiment, computer vision model may process ultrasound images 1104, to make a determination about a scene, space, and / or object in

[0339] 99 Attorney Docket No. 1518-193PCT1 heart. In a non-limiting example, computer vision model may be used for registration of plurality of voxels within a 3D space. In some cases, registration may include image processing described herein, such as without limitation object recognition, feature detection, edge / corner detection, and the like. Non-limiting example of feature detection may include scale invariant feature transform (SIFT), Canny edge detection, Shi Tomasi corner detection, and the like. In some cases, registration may include one or more transformations to orient an ultrasonic image relative a 3D coordinate system; exemplary transformations include without limitation, homography transforms and affine transforms. In an embodiment, registration of ultrasonic image to a coordinate system may be verified and / or corrected using object identification and / or computer vision, as described above. For instance, and without limitation, an initial registration to two dimensions, represented for instance as registration to the x and y coordinates, may be performed using a two-dimensional projection of points in three dimensions onto the ultrasonic image; however, a third dimension of registration, representing depth and / or a z axis, may be detected by utilizing depth-sensing techniques such as Doppler imaging. Alternatively, the third dimension may be inferred from the known geometry and orientation of the imaging device (e.g., ultrasound sensor 1106 of TEE system 1102), or through the application of one or more machine learning models trained to interpret depth from the two-dimensional projection.

[0340] With continued reference to FIG. 11, processor 1110 may use a machine learning module to implement one or more algorithms or generate one or more machine learning models, such as an cardiac modeling model to generate 3D cardiac model 1137 of heart. However, the machine learning module is exemplary and may not be necessary to generate one or more machine learning models and perform any machine learning described herein. In one or more embodiments, one or more machine-learning models may be generated using training data. Training data may include inputs and corresponding predetermined outputs so that a machinelearning model may use correlations between the provided exemplary inputs and outputs to develop an algorithm and / or relationship that then allows machine-learning model to determine its own outputs for inputs. Training data may contain correlations that a machine-learning process may use to model relationships between two or more categories of data elements. Exemplary inputs and outputs may come from a database, such as any database described in this disclosure, or be provided by a user. In other embodiments, a machine-learning module may obtain a training set by querying a communicatively connected database that includes past inputs

[0341] 100 Attorney Docket No. 1518-193PCT1 and outputs. Training data may include inputs from various types of databases, resources, and / or user inputs and outputs correlated to each of those inputs so that a machine-learning model may determine an output. Correlations may indicate causative and / or predictive links between data, which may be modeled as relationships, such as mathematical relationships, by machine-learning models, as described in further detail below. In one or more embodiments, training data may be formatted and / or organized by categories of data elements by, for example, associating data elements with one or more descriptors corresponding to categories of data elements. As a nonlimiting example, training data may include data entered in standardized forms by persons or processes, such that entry of a given data element in a given field in a form may be mapped to one or more descriptors of categories. Elements in training data may be linked to descriptors of categories by tags, tokens, or other data elements. In a further embodiment, training data may include previous outputs such that one or more machine learning models iteratively produces outputs.

[0342] Still referring to FIG. 11, machine learning module may be used to generate anatomy modeling model and / or any other machine learning models, such as, shape identification model as described in further detail below, using training data. Cardiac modeling model may be trained by correlated inputs and outputs of training data. Training data may be data sets that have already been converted from raw data whether manually, by machine, or any other method. In an embodiment, generating 3D cardiac model 1137 of heart may include receiving anatomy training data, wherein the anatomy training data may include a plurality of ultrasound images as input and a plurality of 3D cardiac models as output. In some cases, anatomy training data may be received from LAAC database 1116 or other databases. In other cases, anatomy training data may be collected by a data acquisition unit from external sources such as one or more medical equipment’s e.g., imaging devices or diagnostic tools, wherein the data acquisition may be configured as an intermediary between the data source and machine learning module. In one or more embodiments, anatomy training data may include a plurality of ultrasound images correlated to a plurality of 3D cardiac models. In one or more embodiments, a particular ultrasound images 1104 within anatomy training data may be correlated to a particular 3D cardiac model. In one or more embodiments, anatomy training data may further include a plurality of ultrasound images 1104 correlated to a plurality of cardiac models. In an embodiment, a particular ultrasound images 1104 may be correlated to a particular cardiac

[0343] 101 Attorney Docket No. 1518-193PCT1 model. In one or more embodiments anatomy training data may include TEE diagrams, Cardiac CTs, ECG signals and / or ultrasonic images as an input and correlated 3D representations of heart

[0344] With continued reference to FIG. 11, in an embodiment, anatomy modeling model may include a deep neural network (DNN). As used in this disclosure, a “deep neural network” is defined as a neural network with two or more hidden layers. Neural network is described in further detail below with reference to FIGS. 7 - 9. In a non-limiting example, anatomy modeling model may include a convolutional neural network (CNN). Generating 3D cardiac model 1137 of heart may include training CNN using anatomy training data and generating 3D cardiac model 1137 as a function of ultrasound images 1104 using trained CNN. A “convolutional neural network,” for the purpose of this disclosure, is a neural network in which at least one hidden layer is a convolutional layer that convolves inputs to that layer with a subset of inputs known as a “kernel,” along with one or more additional layers such as pooling layers, fully connected layers, and the like. In some cases, CNN may include, without limitation, a deep neural network (DNN) extension. Mathematical (or convolution) operations performed in the convolutional layer may include convolution of two or more functions, where the kernel may be applied to input data e.g., ultrasound images 1104 through a sliding window approach. In some cases, convolution operations may enable processor 1110 to detect local / global patterns, edges, textures, and any other cardiac feature datums 1138 described herein within each ultrasound images 1104. Cardiac feature datums 1138 may be passed through one or more activation functions, such as without limitation, Rectified Linear Unit (ReLU), to introduce non-linearities into the processing step of generating 3D cardiac model 1137 of heart. Additionally, or alternatively, CNN may also include one or more pooling layers, wherein each pooling layer is configured to reduce the dimensionality of input data while preserving essential features within the input data. In a non-limiting example, CNN may include one or more pooling layer configured to reduce the spatial dimensions of cardiac feature datum maps by applying downsampling, such as max-pooling or average pooling, to small, non-overlapping regions of one or more cardiac feature datums 1138.

[0345] Still referring to FIG. 11, CNN may further include one or more fully connected layers configured to combine cardiac feature datums 1138 extracted by the convolutional and pooling layers. In some cases, one or more fully connected layers may allow for higher-level pattern recognition. In a non-limiting example, one or more fully connected layers may connect every

[0346] 102 Attorney Docket No. 1518-193PCT1 neuron (i.e., node) in its input to every neuron in its output, functioning as a traditional feedforward neural network layer. In some cases, one or more fully connected layers may be used at the end of CNN to perform high-level reasoning and produce the final output such as, without limitation, a 3D cardiac model 1137 of heart. Further, each fully connected layer may be followed by one or more dropout layers configured to prevent overfitting, and one or more normalization layers to stabilize the learning process described herein.

[0347] With continued reference to FIG. 11, CNN may further include a 3D CNN, wherein the 3D CNN, unlike standard 2D CNN, may include utilization of one or more 3D convolutions which allow them to directly process 3D data, thereby enabling processor 1110 to generate 3D structures such as 3D cardiac model 1137 of heart using the 3D CNN. In a non-limiting example, 3D CNN may include one or more 3D filters (i.e., kernels) that move through the ultrasound images 1104 in three dimensions and capturing spatial relationships in x, y, and z axis. Similar to 3D convolutions, 3D CNN may further include one or more 3D pooling layers that may be used to reduce the dimensionality of ultrasonic images while preserving cardiac feature datums 1138 as described above. Additionally, or alternatively, an encoder-decoder structure may be implemented (extended to 3D), by processor 1110, in 3D CNN, wherein the encoder-decoder structure includes an encoding path that captures the context and a decoding path that enables precise localization in a same manner as U-net as described above. Such encoder-decoder structures may also include a plurality of skip connections, allowing 3D CNN to use information from multiple resolutions to improve the process of generating 3D cardiac model 1137 of heart.

[0348] With continued reference to FIG. 11, in an embodiment, training the cardiac modeling model (i.e., CNN) may include selecting a suitable loss function to guide the training process. In a non-limiting example, a loss function that measures the difference between the predicted 3D VORs and the ground truth 3D structure e g., CT-based anatomical object models may be used, such as, without limitation, mean squared error (MSE) or a custom loss function may be designed for one or more embodiments described herein. Additionally, or alternatively, optimization algorithms, such as stochastic gradient descent (SGD), may then be used to adjust the anatomy modeling model’s parameters to minimize such loss. In a further non-limiting embodiment, instead of directly predicting 3D cardiac model 1137, cardiac modeling model may be trained as a regression model to predict embedded values described herein for each voxel of plurality of voxels within a 3D grid. Additionally, CNN may be extended with additional deep

[0349] 103 Attorney Docket No. 1518-193PCT1 learning techniques, such as recurrent neural networks (RNNs) or attention mechanism, to capture additional features and / or data relationships within input data. These extensions may further enhance the accuracy and robustness of the anatomical object modeling.

[0350] With continued reference to FIG. 11, in some embodiments, generating at least a 3D cardiac model 1137 may include extracting at least a cardiac feature datum 1138 from at least an ultrasound image 1104 and segmenting the at least an ultrasound image 1104 into a plurality of image segments 1142. As a non-limiting example, extracting cardiac feature datum 1138 may include structural elements like left atrium, left atrial appendage, mitral valve, aortic valve, ventricular walls, and the like. In some embodiments, extracting a cardiac feature datum 1138 may include isolating and identifying cardiac feature datum 1138 from ultrasound image 1104 using image processing or machine learning techniques. For the purposes of this disclosure, an “image segment” is a section of an ultrasound image. In some embodiments, image segment 1142 may include a section of an ultrasound image that has been partitioned based on shared visual or anatomical properties (e.g., cardiac feature datum 1138). In a non-limiting example, in a 2D image, segmenting ultrasound image 1104 may include all the pixels that make up cardiac feature datum 1138, while in a 3D context, it would encompass all the voxels (3D pixels) that constitute cardiac feature datum 1138. In some embodiments, processor 1110 may segment 3D cardiac model 1137 and segmenting 3D cardiac model 1137 may include extracting cardiac feature datum 1138 from 3D cardiac model 1137 and segmenting 3D cardiac model 1137 as a function of cardiac feature datum 1138.

[0351] With continued reference to FIG. 11, in some embodiments, processor 1110 may be configured to analyze ultrasound image and / or 3D cardiac model 1137 using machine vision system to determine cardiac feature datum 1138. For the purposes of this disclosure, a “machine vision system” is a type of technology that enables a computing device to inspect, evaluate and identify still or moving images. For example, in some cases a machine vision system may be used for world modeling or registration of objects within a space. In some cases, registration may include image processing, such as without limitation object recognition, feature detection, edge / comer detection, and the like. Non-limiting example of feature detection may include scale invariant feature transform (SIFT), Canny edge detection, Shi Tomasi corner detection, and the like. In some cases, a machine vision process may operate image classification and segmentation models, such as without limitation by way of machine vision resource (e.g., OpenMV or

[0352] 104 Attorney Docket No. 1518-193PCT1 TensorFlow Lite). A machine vision process may detect motion, for example by way of frame differencing algorithms. A machine vision process may detect markers, for example blob detection, object detection (e.g., cardiac feature datum 1138), face detection, and the like.

[0353] With continued reference to FIG. 11, in some cases, registration may include one or more transformations to orient a camera frame (or an image or video stream) relative a three- dimensional coordinate system; exemplary transformations include without limitation homography transforms and affine transforms. In an embodiment, registration of first frame to a coordinate system may be verified and / or corrected using object identification and / or computer vision, as described above. For instance, and without limitation, an initial registration to two dimensions, represented for instance as registration to the x and y coordinates, may be performed using a two-dimensional projection of points in three dimensions onto a first frame, however. A third dimension of registration, representing depth and / or a z axis, may be detected by comparison of two frames; for instance, where first frame includes a pair of frames captured using a pair of cameras (e.g., stereoscopic camera also referred to in this disclosure as stereocamera), image recognition and / or edge detection software may be used to detect a pair of stereoscopic views of images of an object; two stereoscopic views may be compared to derive z- axis values of points on object permitting, for instance, derivation of further z-axis points within and / or around the object using interpolation. This may be repeated with multiple objects in field of view, including without limitation environmental features of interest identified by object classifier and / or indicated by an operator. In an embodiment, x and y axes may be chosen to span a plane common to two cameras used for stereoscopic image capturing and / or an xy plane of a first frame; a result, x and y translational components and (f> may be pre-populated in translational and rotational matrices, for affine transformation of coordinates of object, also as described above. Initial x and y coordinates and / or guesses at transformational matrices may alternatively or additionally be performed between first frame and second frame, as described above. For each point of a plurality of points on object and / or edge and / or edges of object as described above, x and y coordinates of a first stereoscopic frame may be populated, with an initial estimate of z coordinates based, for instance, on assumptions about object, such as an assumption that ground is substantially parallel to an xy plane as selected above. Z coordinates, and / or x, y, and z coordinates, registered using image capturing and / or object identification processes as described above may then be compared to coordinates predicted using initial guess

[0354] 105 Attorney Docket No. 1518-193PCT1 at transformation matrices; an error function may be computed using by comparing the two sets of points, and new x, y, and / or z coordinates, may be iteratively estimated and compared until the error function drops below a threshold level.

[0355] With continued reference to FIG. 11, in some embodiments, segmenting a ultrasound image 1104 and / or 3D cardiac model 1137 may include training a segmentation model with segmentation training data, wherein the segmentation training data may include exemplary ultrasound image and / or 3D cardiac model correlated to exemplary segmented ultrasound image and / or 3D cardiac model and segmenting a ultrasound image 1104 and / or 3D cardiac model 1137 using the trained segmentation model. For the purposes of this disclosure, a “segmentation model” is a machine learning or deep learning model designed to partition an image into multiple segments or regions, each corresponding to different objects or parts of an object within the image. In some embodiments, segmentation model may assign a label to each pixel in an image (e g., ultrasound image 1104 and / or 3D cardiac model 1137) such that pixels with the same label share certain characteristics, such as belonging to the same cardiac feature datum 1138 or region. As a non-limiting example, segmentation model may include a neural network. For the purposes of this disclosure, “segmentation training data” is data containing correlations that a machinelearning process may use to model relationships between echo depth maps and segmented echo depth maps. In a non-limiting example, a segmentation model may analyze ultrasound image 1104 and / or 3D cardiac model 1137 to identify and delineate the boundaries of cardiac feature datum 1138. This may include finding the set of coordinates {(X), F;)} that represent the pixels or voxels making up the cardiac feature datum 1138. In some embodiments, processor 1110 may segment ultrasound image and / or 3D cardiac model 1137 based on cardiac feature datum 1138. In some embodiments, segmentation training data may be stored in LAAC database 1116. In some embodiments, segmentation training data may be received from one or more users, LAAC database 1116, external computing devices, and / or previous iterations of processing. As a nonlimiting example, segmentation training data may include instructions from a user, who may be an expert user, a past user in embodiments disclosed herein, or the like, which may be stored in memory and / or stored in LAAC database 1116, where the instructions may include labeling of training examples. In some embodiments, segmentation training data may be updated iteratively on a feedback loop. As a non-limiting example, processor 1110 may update segmentation training data iteratively through a feedback loop as a function of output of feature extraction

[0356] 106 Attorney Docket No. 1518-193PCT1 model, ultrasound image 1104, 3D cardiac model 1 137, and the like. In some embodiments, processor 1110 may be configured to generate segmentation model. In a non-limiting example, generating segmentation model may include training, retraining, or fine-tuning segmentation model using segmentation training data or updated segmentation training data. In some embodiments, processor 1110 may be configured to segment ultrasound image 1104 and / or 3D cardiac model 1137 using segmentation model (i.e., trained or updated segmentation model).

[0357] With continued reference to FIG. 11, in some embodiments, generating at least a 3D cardiac model 1137 may include generating a 3D point cloud 1140 as a function of a plurality of image segments 1142 and generating a 3D mesh model 1144 of at least a 3D cardiac model 1137 as a fun...

Claims

WHAT IS CLAIMED IS:

1. An apparatus for determining a cardiac implant size, the apparatus comprising: an ultrasonic imaging device; at least a processor; and a memory communicatively connected to the at least a processor, wherein the memory contains instructions configuring the at least a processor to: using the ultrasonic imaging device, collect a plurality of ultrasonic images; using a 3D cardiac model generation machine learning model trained on a training dataset comprising example ultrasonic images correlated with example 3D cardiac models, generate a 3D cardiac model based on the plurality of ultrasonic images; generate at least a cardiac measurement based on the 3D cardiac model; determine a cardiac implant size based on the at least a cardiac measurement; and using a display, display to a user the cardiac implant size.

2. The apparatus of claim 1, wherein the cardiac implant size comprises a left atrial appendage occlusion device size.

3. The apparatus of claim 1, wherein the memory contains instructions further configuring the at least a processor to determine one or both of a cardiac implant placement and a cardiac implant candidate as a function of the 3D cardiac model.

4. The apparatus of claim 1, wherein the memory contains instructions further configuring the at least a processor to: receive at least an ultrasound localization datum representing one or both of position and angle of the ultrasonic imaging device; and generate, using the 3D cardiac model generation machine learning model, the 3D cardiac model as a function of the at least an ultrasound localization datum.

5. The apparatus of claim 1, wherein the plurality of ultrasonic images comprises one or more of transesophageal echocardiogram (TEE) images, intracardiac echocardiogram (ICE) images, transthoracic echocardiogram (TTE) images, and point of care ultrasound (POCUS) images.

6. The apparatus of claim 1, wherein the memory contains instructions further configuring the at least a processor to:126 Attorney Docket No. 1518-193PCT1calculate a level of uncertainty at a plurality of locations on the 3D cardiac model, wherein the plurality of locations comprises a high uncertainty region; receive a subsequent plurality of ultrasonic images of cardiac anatomy corresponding to a high uncertainty region of the 3D cardiac model, wherein the subsequent plurality of ultrasonic images is captured using the ultrasonic imaging device, as a function of the high uncertainty region; and generate a subsequent 3D cardiac model as a function of the subsequent plurality of ultrasonic images.

7. The apparatus of claim 1, wherein generating the 3D cardiac model based on the plurality of ultrasonic images comprises: generating a set of shape parameters representing a cardiac shape as a function of the plurality of ultrasonic images and a shape identification model trained on the training dataset; wherein the set of shape parameters comprises a plurality of numerical descriptors representing at least a geometric characteristic of a subject’s heart and a plurality of associated parameter ranges.

8. The apparatus of claim 1, wherein the memory contains instructions further configuring the at least a processor to: receive a 3D cardiac implant model representing a cardiac implant; and display, using the display, the 3D cardiac implant model positioned within the 3D cardiac model.

9. The apparatus of claim 1, wherein the cardiac measurement comprises an ostial diameter of a left atrial appendage (LAA).

10. The apparatus of claim 1, wherein generating the 3D cardiac model based on the plurality of ultrasonic images comprises: generating a 3D voxel occupancy representation (VOR) representing a cardiac shape as a function of the plurality of ultrasonic images and the 3D cardiac model generation machine learning model trained on the training dataset; generating a mesh representing the cardiac shape as a function of the 3d voxel occupancy representation; and displaying, using the display, the mesh to the user.127 Attorney Docket No. 1518-193PCT111. A method of determining a cardiac implant size, the method comprising: using at least a processor and an ultrasonic imaging device, collecting a plurality of ultrasonic images; using the at least a processor and a 3D cardiac model generation machine learning model trained on a training dataset comprising example ultrasonic images correlated with example 3D cardiac models, generating a 3D cardiac model based on the plurality of ultrasonic images; using the at least a processor, generating at least a cardiac measurement based on the 3D cardiac model; using the at least a processor, determining a cardiac implant size based on the at least a cardiac measurement; and using the at least a processor and a display, displaying to a user the cardiac implant size.

12. The method of claim 11, wherein the cardiac implant size comprises a left atrial appendage occlusion device size.

13. The method of claim 11, wherein the method further comprises determining one or both of a cardiac implant placement and a cardiac implant candidate quality as a function of the 3D cardiac model.

14. The method of claim 11, wherein the method further comprises: receiving, using the at least a processor, at least an ultrasound localization datum representing one or both of position and angle of the ultrasonic imaging device; and generating, using the at least a processor and the 3D cardiac model generation machine learning model, the 3D cardiac model as a function of the at least an ultrasound localization datum.

15. The method of claim 11, wherein the plurality of ultrasonic images comprises one or more of transesophageal echocardiogram (TEE) images, intracardiac echocardiogram (ICE) images, transthoracic echocardiogram (TTE) images, and point of care ultrasound (POCUS) images.

16. The method of claim 11, wherein the method further comprises: calculating, using the at least a processor, a level of uncertainty at a plurality of locations on the 3D cardiac model, wherein the plurality of locations comprises a high128 Attorney Docket No. 1518-193PCT1uncertainty region; receiving, using the at least a processor, a subsequent plurality of ultrasonic images of cardiac anatomy corresponding to a high uncertainty region of the 3D cardiac model, wherein the subsequent plurality of ultrasonic images is captured using the ultrasonic imaging device, as a function of the high uncertainty region; and generating, using the at least a processor, a subsequent 3D cardiac model as a function of the subsequent plurality of ultrasonic images.

17. The method of claim 11, wherein generating the 3D cardiac model based on the plurality of ultrasonic images comprises: generating a set of shape parameters representing a cardiac shape as a function of the plurality of ultrasonic images and a shape identification model trained on the training dataset; wherein the set of shape parameters comprises a plurality of numerical descriptors representing at least a geometric characteristic of a subject’s heart and a plurality of associated parameter ranges.

18. The method of claim 11, wherein the method further comprises: receiving, using the at least a processor, a 3D cardiac implant model representing a cardiac implant; and displaying, using the at least a processor and the display, the 3D cardiac implant model positioned within the 3D cardiac model.

19. The method of claim 11, wherein the cardiac measurement comprises an ostial diameter of a left atrial appendage (LAA).

20. The method of claim 11, wherein the generating the 3D cardiac model based on the plurality of ultrasonic images comprises: generating, using the at least a processor, a 3D voxel occupancy representation (VOR) representing a cardiac shape as a function of the plurality of ultrasonic images and the 3D cardiac model generation machine learning model trained on the training dataset; generating, using the at least a processor, a mesh representing the cardiac shape as a function of the 3D voxel occupancy representation; and displaying, using the at least a processor and the display, the mesh to a user.129 Attorney Docket No. 1518-193PCT121 . A system for transesophageal echocardiogram-guided implantation of a left atrial appendage closure device, the system comprising: at least a transesophageal echocardiogram (TEE) system comprising at least an ultrasound sensor, wherein the at least an ultrasound sensor is configured to be located within an esophagus of a patient and detect at least an ultrasound image as a function of cardiac tissue of the patient; at least a display; and at least a computing device comprising at least a processor and a memory containing instructions configuring the at least a processor to: receive the at least an ultrasound image; generate at least a three-dimensional (3D) cardiac model representative of a heart of the patient as a function of the at least an ultrasound image; receive at least a 3D left atrial appendage closure (LAAC) device model representative of at least a LAAC device; generate a superimposed model by superimposing the at least a 3D LAAC device model onto the at least a 3D cardiac model; and display, using the at least a display, the superimposed model.

22. The system of claim 21, wherein receiving the at least an ultrasound image comprises: extracting at least a TEE angle datum from the at least an ultrasound image using an optical character recognition; and generating an image inquiry datum as a function of the at least a TEE angle datum and device instruction for use (IFU) data retrieved from an LAAC database, wherein the image inquiry datum is configured to query additional ultrasound images to a user of the at least a TEE system through the at least a display.

23. The system of claim 21, wherein receiving the at least an ultrasound image comprises: generating view training data, wherein the view training data comprises exemplary ultrasound images correlated to exemplary view labels; training a view classifier using the view training data; classifying the at least an ultrasound image to at least a view label using the trained view classifier; and generating an image inquiry datum as a function of the view label and IFU data.130 Attorney Docket No. 1518-193PCT124. The system of claim 21, wherein generating the at least a 3D cardiac model comprises: extracting at least a cardiac feature from the at least an ultrasound image; and segmenting the at least an ultrasound image into a plurality of image segments.

25. The system of claim 24, wherein generating the at least a 3D cardiac model comprises: generating a 3D point cloud as a function of the plurality of image segments; and generating a 3D mesh model of the at least a 3D cardiac model as a function of the 3D point cloud.

26. The system of claim 21, wherein generating the at least a 3D cardiac model comprises generating the at least a 3D cardiac model using a statistical shape model.

27. The system of claim 21, wherein receiving the at least a 3D LAAC device model comprises: extracting at least an ostium characteristic datum from the at least an ultrasound image; determining a device datum as a function of the at least an ostium characteristic datum and a compression rate of a plurality of LAAC devices, wherein the device datum comprises a size datum; and generating the at least a 3D LAAC device model as a function of the device datum.

28. The system of claim 27, wherein determining the device datum comprises: simulating a placement of the plurality of LAAC devices within the at least a 3D cardiac model as a function of the at least an ostium characteristic datum and the compression rate; and generating the device datum as a function of the simulation.

29. The system of claim 27, wherein determining the device datum comprises determining a pass datum as a function of the at least an ostium characteristic datum and the compression rate.

30. The system of claim 21, wherein superimposing the at least a 3D LAAC device model onto the at least a 3D cardiac model comprises: determining an optimal path for a placement of the at least a 3D LAAC device model within the at least a 3D cardiac model; generating a path model for the optimal path; and superimposing the path model onto the at least a 3D cardiac model.

31. The system of claim 21 , wherein the 3D cardiac model comprises peripheral vasculature.131 Attorney Docket No. 1518-193PCT132. The system of claim 21, wherein generating the at least a 3D cardiac model comprises generating the at least a 3D cardiac model using a point completion model.

33. A method for transesophageal echocardiogram -guided implantation of a left atrial appendage closure device, the method comprising: receiving, using at least a processor, at least an ultrasound image from at least a transesophageal echocardiogram (TEE) system comprising at least an ultrasound sensor, wherein the at least an ultrasound sensor is configured to be located within an esophagus of a patient and detect the at least an ultrasound image as a function of cardiac tissue of the patient; generating, using the at least a processor, at least a three-dimensional (3D) cardiac model representative of a heart of the patient as a function of the at least an ultrasound image; receiving, using the at least a processor, at least a 3D left atrial appendage closure (LAAC) device model representative of at least a LAAC device; generating, using the at least a processor, a superimposed model by superimposing the at least a 3D LAAC device model onto the at least a 3D cardiac model; and displaying, using the at least a processor and at least a display, the superimposed model.

34. The method of claim 33, wherein receiving the at least an ultrasound image comprises: extracting at least a TEE angle datum from the at least an ultrasound image using an optical character recognition; and generating an image inquiry datum as a function of the at least a TEE angle datum and device instruction for use (IFU) data retrieved from an LAAC database, wherein the image inquiry datum is configured to query additional ultrasound images to a user of the at least a TEE system through the at least a display.

35. The method of claim 33, wherein receiving the at least an ultrasound image comprises: generating view training data, wherein the view training data comprises exemplary ultrasound images correlated to exemplary view labels; training a view classifier using the view training data; classifying the at least an ultrasound image to at least a view label using the trained view classifier; and generating an image inquiry datum as a function of the view label and IFU data.132 Attorney Docket No. 1518-193PCT136. The method of claim 33, wherein generating the at least a 3D cardiac model comprises: extracting at least a cardiac feature from the at least an ultrasound image; and segmenting the at least an ultrasound image into a plurality of image segments.

37. The method of claim 36, wherein generating the at least a 3D cardiac model comprises: generating a 3D point cloud as a function of the plurality of image segments; and generating a 3D mesh model of the at least a 3D cardiac model as a function of the 3D point cloud.

38. The method of claim 33, wherein generating the at least a 3D cardiac model comprises generating the at least a 3D cardiac model using a statistical shape model.

39. The method of claim 33, wherein receiving the at least a 3D LAAC device model comprises: extracting at least an ostium characteristic datum from the at least an ultrasound image; determining a device datum as a function of the at least an ostium characteristic datum and a compression rate of a plurality of LAAC devices, wherein the device datum comprises a size datum; and generating the at least a 3D LAAC device model as a function of the device datum.

40. The method of claim 39, wherein determining the device datum comprises: simulating a placement of the plurality of LAAC devices within the at least a 3D cardiac model as a function of the at least an ostium characteristic datum and the compression rate; and generating the device datum as a function of the simulation.

41. The method of claim 39, wherein determining the device datum comprises determining a pass datum as a function of the at least an ostium characteristic datum and the compression rate.

42. The method of claim 41, wherein superimposing the at least a 3D LAAC device model onto the at least a 3D cardiac model comprises: determining an optimal path for a placement of the at least a 3D LAAC device model within the at least a 3D cardiac model; generating a path model for the optimal path; and superimposing the path model onto the at least a 3D cardiac model.

43. The method of claim 33, wherein the 3D cardiac model comprises peripheral vasculature.133 Attorney Docket No. 1518-193PCT144. The method of claim 33, wherein generating the at least a 3D cardiac model comprises generating the at least a 3D cardiac model using a point completion model.

45. A system for transesophageal echocardiogram -guided implantation of a left atrial appendage closure device, the system comprising: at least a transesophageal echocardiogram system comprising at least an ultrasound sensor configured to be located within an esophagus of a patient and detect at least an ultrasound image as a function of cardiac tissue of the patient; at least a display; and at least a computing device configured to: receive the at least an ultrasound image; input the at least an ultrasound image into a 3D reconstruction model; generate, using the 3D reconstruction model, at least a 3D cardiac model representative of a patient’s heart, as a function of the at least an ultrasound image; receive at least a 3D watchman model representative of at least a left atrial appendage closure device; and display, using the at least a display, the at least an ultrasound image, the at least a 3D cardiac model, and the at least a 3D watchman model.

46. A method of using the system of claim 45 to plan implantation of a left atrial appendage closure device, comprising: providing feedback to an operator to ensure that the at least an ultrasound sensor has been position and orientation to capture frames and views that are in sync with a Watchman IFU.

47. The method of claim 46, further comprising: determining, using a placement and size determination engine, one or more of placement, including position and orientation, and size of a watchman device.

48. A method of using the system of claim 45 to guide implantation of a left atrial appendage closure device.

49. A method of using the system of claim 45 post-implantation of a left atrial appendage closure device, comprising: visualizing an actual implanted watchman device, using a post-implantation 3D reconstruction model.134 Attorney Docket No. 1518-193PCT10. The method of claim 49 further comprising displaying, using the at least a display, a 3D constructed image of the actual implanted watchman device and the 3D watchman model overlaid, with recommended watchman placement.135 Attorney Docket No. 1518-193PCT1