Systems and methods for lesion detection

Abstract

The invention relates to systems and methods for lesion detection based on heart tissue dynamics determined via one or more algorithms for analyzing ultrasound image data associated with an anatomical region of interest. In particular, systems of the invention provide for processing the ultrasound image data to reconstruct images of the tissue representing the anatomical region of interest during a complete single heartbeat and to analyze a stress to strain relationship of the tissue to identify areas of one or more lesions in the tissue.

Description

[0001] SYSTEMS AND METHODS FOR LESION DETECTION

[0002] Cross-Reference to Related Applications

[0003] This application claims priority to, and the benefit of, U.S. Provisional Application No. 63 / 666,438, filed July 1, 2024, the content of which is incorporated by reference herein in its entirety.

[0004] Field of the Invention

[0005] The invention generally relates to ultrasound imaging, and, more particularly, to systems and methods for lesion detection based on heart tissue dynamics.

[0006] Background

[0007] Atrial fibrillation (AF) as well as other complex cardiac arrythmias such as atrial flutter (AFL) or ventricular tachycardia (VT) are defined through irregular heartbeats caused by chaotic electrical signals in the atrial or ventricular chambers of the heart. Currently, it is estimated that more than 33 million individuals worldwide have AF, which is associated with many adverse outcomes, including stroke, dementia, heart failure, impaired quality of life, and increased medical costs. AF increases the risk of stroke by an average of 5-fold, and AF-related strokes are more severe than those not related to AF. AF causes a wide variety of symptoms, including fatigue and reduced exercise tolerance. Further, in the United States, AF accounts for more than 450,000 hospitalizations yearly, and is reported to increase annual health care costs by $8700 per patient, resulting in a $26 billion annual increase in U.S. health care costs.

[0008] Cardiac ablation is a common treatment approach for arrythmias, where specific regions within the heart are destroyed through ablation. In cardiac ablation, scars or lesions are created in cardia tissue, for example via RF energy, pulsed field ablation (PF A), laser, or cryotherapy, to prevent or interrupt the transmission of abnormal electrical signals. Cardiac ablation forms an essential part of the management of cardiac arrhythmias, including supraventricular tachycardia, atrial flutter, atrial fibrillation, and ventricular tachycardia.

[0009] For interventional cardiac applications, such as cardiac ablation, accurately capturing a visual representation of the anatomy of interest is paramount for a successful procedure. Successful catheter ablation requires not only precise localization of the arrhythmogenic substrate, but complete and permanent elimination of that substrate without producing collateral injury. The ablation effect depends on a number of factors, including applied electrical power, quality of the tissue contact, local tissue properties, presence of blood flow close to the tissue surface, and the effect of irrigation. Because of the variability of these parameters, it is difficult to obtain consistent results and to understand ablation effects in tissue using current systems and methods for ablation.

[0010] Notably, despite extensive utilization of imaging equipment and tools for ablation, as well as systems supporting the identification of electrically active regions of the heart, AF treatment still has a relatively low efficacy. The relatively low efficacy of AF treatment is likely due to limitations in mapping, incomplete understanding of the driving mechanisms of arrhythmia, and, most importantly, the inability to create transmural and durable lesions. As a result, there exists a need for improved imaging to diagnose and treat arrhythmia and / or abnormal heart rhythms.

[0011] Summary

[0012] The present invention addresses the limitations of currently utilized approaches for treating arrhythmias using interventional cardiac echography (ICE) ultrasound systems and provides improved systems and methods for lesion detection and analysis. In particular, the invention provides systems and methods for differentiating untreated cardiac tissue from ablated cardiac tissue during electrophysiology (EP) therapy and / or during cardiac ablation. Further, the systems and methods of the invention provide for analyzing cardiac heart function within a single beat to determine the stress and strain relationship of cardiac tissue and to identify areas of altered strain within the tissue. By analyzing areas of altered strain within the tissue, the invention provides for both identifying a lesion and for measuring an ablation depth profile in the tissue over the ablation path.

[0013] The systems and methods of the invention provide for acquisition of two-dimensional (2D) and three-dimensional (3D) volume (2D / 3D volume) ultrasound data. For example data may be acquired from a cylindrical array with a matrix array of transducer elements in rows over the complete perimeter of the catheter tip, and / or any rotating transducer array system capable of real-time 2D or 3D ultrasound imaging. Further, the systems and methods of the invention use machine learning approaches for the analysis of the stress and strain of heart movement in all three-dimensional directions, in combination with pattern-based tracking in both 2D images and 3D volumes, to provide real-time imaging and analysis of a tissue region of interest. The imaging and analysis allows the clinician to differentiate between ablated tissue and untreated tissue and thus identifies lesion location and depth of the ablated tissue.

[0014] The invention recognizes that, despite the extensive utilization of imaging equipment and tools for ablation, as well as systems supporting the identification of electrically active regions within the heart (e.g. electroanatomical mapping systems), conventional imaging systems and methods do not allow a clinical user to understand the complex endocardial and myocardial anatomy live and in sufficient detail to provide a clinician with highly reliable anatomical and physiological feedback before, during, and / or after an ablation procedure. The systems and methods of the invention address this problem and provide for fast, accurate, and live lesion detection and analysis through analyzing heart tissue dynamics in combination with machinelearning techniques. Accordingly, the invention provides for intra-procedural feedback to enable patient-specific planning and ablation monitoring compatible with any ablation techniques, for example, thermal ablation, radio frequency (RF) ablation, pulsed field ablation (PF A) cryoablation, laser ablation, and ultrasound procedures.

[0015] Aspects of the invention provide systems for analyzing image data associated with cardiac tissue. The systems include a hardware processor coupled to non-transitory, computer- readable memory containing instructions executable by the processor. The instructions executable by the processor cause the processor to receive ultrasound image data associated with an anatomical area of interest, and run a tissue analysis algorithm to identify one or more lesions present in tissue at the anatomical area of interest. Running the tissue analysis algorithm includes processing the data to generate one or more images of tissue at the anatomical area of interest, wherein the one or more images represent cardiac tissue during a complete single heartbeat. Running the tissue analysis algorithm further includes analyzing a stress to strain relationship of the tissue, and identifying altered strain in the tissue and detecting one or more lesions in the tissue based, at least in part, on identified altered strain.

[0016] In some embodiments, analyzing the stress to strain relationship comprises calculating a stress induced through a displacement of the tissue and a resulting strain of the tissue using one or more of a pattern-based tracking method, a block matching method, and / or tracking of one or more anatomical regions comprising one or more anatomical landmarks and / or structures, for example endocardial and epicardial walls, or pulmonary veins. Further, in some embodiments, the images are one or more of two-dimensional (2D) and three-dimensional (3D) volume (2D / 3D volume) images. In particular embodiments, the pattern-based method comprises a speckle tracking analysis within the 2D / 3D volume, wherein a plurality of ultrasound speckles from the 2D / 3D volume are used as input for calculating the stress and the strain.

[0017] In some embodiments, the complete single heartbeat is divided into discrete time points such that an individual 2D / 3D volume represents a separate time point within the complete single heartbeat. For example, in some embodiments, the algorithm calculates the displacement of tissue and the resulting strain based on a movement of the plurality of speckles between time points.

[0018] In some embodiments, a displacement between consecutive 2D / 3D volumes are used to calculate the stress and the strain of the tissue. Further, a time point when the stress of the tissue is a minimum or a time point when the stress of the tissue is a maximum is calculated from a contraction pattern, wherein the contraction pattern is identified via an electrocardiogram (ECG) or an image based method, in some embodiments.

[0019] In some embodiments, the algorithm is a machine learning algorithm trained on a plurality of 2D / 3D volume data comprising a plurality of identified tissue displacement in a plurality of complete single heartbeats, wherein a plurality of strongest displacements in the 2D / 3D volume data are identified. Further, the machine learning algorithm comprises a neural network-based feature extractor to predict the tissue displacement from a first 2D / 3D volume to a second 2D / 3D volume, in some embodiments. In particular embodiments, the feature extractor uses speckle patterns, and / or one or more other patterns that are distinctive for a region of interest, wherein the speckle patterns and / or the one or more other patterns appear in both the first 2D / 3D volume and the second 2D / 3D volume. Further, the algorithm is further trained on labeled training data for end-to-end prediction of strain from two consecutive 2D / 3D volumes, in some embodiments. In some embodiments, the algorithm is further trained using data comprising a plurality of externally measured strain measurements at a known location in the 2D / 3D volume.

[0020] In some embodiments, identifying altered strain in the tissue differentiates ablated cardiac tissue from untreated cardiac tissue. For example, in some embodiments, the algorithm generates a pre-trained model that predicts strain between two ultrasound volumes. In particular embodiments, the pre-trained model further comprises a segmentation task, wherein the segmentation task differentiates ablated cardiac tissue from untreated tissue. Further, the model is trained using 2D / 3D volumes of different time points from a plurality of heartbeat cycles to achieve a base for identifying differences in strain between 2D / 3D volumes, wherein the model is trained before ablation, some embodiments. For example, in some embodiments, the model is trained during ablation, wherein the model comprises one or more of an unsupervised machine learning model and a supervised model trained with labeled data from one or more of elastography data and histology data identifying lesion tissue. Further, in some embodiments, a difference in strain between 2D / 3D volumes before ablation and during ablation generates a lesion identification, wherein strain differences are predictive of the lesion extent. In particular embodiments, the generated lesion identification is output as a volume overlayed on a model of the ultrasound 2D / 3D volume. In some embodiments, the algorithm calculates an ablation depth to generate an ablation depth profile in the tissue over an ablation path.

[0021] In some embodiments, the ultrasound data is obtained from a ID or 2D matrix array providing 3D imaging with electronic or mechanic beam steering capabilities. In particular embodiments, the received ultrasound image data comprises full circumferential 3D image data obtained from an intracardiac echography (ICE) catheter, wherein the ultrasound imaging data is real-time ultrasound image data comprising 3D volume data associated with cardiac tissue. Further, in some embodiments, the data comprises one or more of a plurality of time-sequentially acquired B-Images, beamformed RF data, and enveloped data over a full perimeter of a catheter tip, wherein the data corresponds to a cross-section of the tissue during the complete single heartbeat. The ultrasound image data comprises a complete simultaneously acquired volume around the full perimeter of a catheter tip, in some embodiments.

[0022] In some embodiments, the system is further configured to preprocess the ultrasound image data, wherein preprocess comprises one or more of noise removal, image smoothing, and contrast enhancement.

[0023] In some embodiments of the systems of the invention, analyzing further comprises analyzing data from one or more ultrasound signals at one or more signal stages, wherein the one or more signal stages comprise compounded, envelope, and log-compressed. In some embodiments, analyzing further comprises analyzing data from one or more methods comprising a Nakagami method, and a microstructural envelop statistics method. In some embodiments, analyzing further comprises analyzing data related to one or more of perfusion, stiffness, anisotropy, coherence, specific statistical distributions in tissue, frequency power spectrum of tissue, ultrasound data at various beamforming stages, one or more learned features from raw and / or processed signals, interventional tool tracking data, and breath data.

[0024] In other aspects, the invention provides methods for analyzing cardiac tissue images. The methods include receiving ultrasound image data associated with an anatomical area of interest, and running a tissue analysis algorithm to identify one or more lesions present in tissue at the anatomical area of interest. Running the tissue analysis algorithm includes processing the data to generate one or more images of tissue at the anatomical area of interest, wherein the one or more images represent cardiac tissue during a complete single heartbeat, analyzing a stress to strain relationship of the tissue, and identifying altered strain in the tissue and detecting one or more lesions in the tissue based, at least in part, on identified altered strain.

[0025] In some embodiments of the methods, analyzing the stress to strain relationship comprises calculating a stress induced through a displacement of the tissue and a resulting strain of the tissue using one or more of a pattern-based tracking method, a block matching method, and / or tracking of one or more anatomical regions comprising one or more anatomical landmarks and / or structures. Further, in some embodiments, the images are one or more of two- dimensional (2D) and three-dimensional (3D) volume (2D / 3D volume) images. The patternbased method comprises a speckle tracking analysis within the 3D volume, wherein a plurality of ultrasound speckles from the 2D / 3D volume are used as input for calculating the stress and the strain, in some embodiments. In some embodiments, the complete single heartbeat is divided into discrete time points such that an individual 2D / 3D volume represents a separate time point within the complete single heartbeat. In particular embodiments, the algorithm calculates the displacement of tissue and the resulting strain based on a movement of the plurality of speckles between time points. Further, in some embodiments a displacement between consecutive 2D / 3D volumes are used to calculate the stress and the strain of the tissue. For example, in some embodiments, a time point when the stress of the tissue is a minimum or a time point when the stress of the tissue is a maximum is calculated from a contraction pattern, wherein the contraction pattern is identified via an electrocardiogram (ECG) or an image-based method.

[0026] In some embodiments of the methods, the algorithm is a machine learning algorithm trained on a plurality of 2D / 3D volume data comprising a plurality of identified tissue displacement in a plurality of complete single heartbeats, wherein a plurality of strongest displacements in the 2D / 3D volume data are identified. The machine learning algorithm comprises a neural network-based feature extractor to predict the tissue displacement from a first 2D / 3D volume to a second 2D / 3D volume, in some embodiments. In particular embodiments, the feature extractor uses speckle patterns, and / or one or more other patterns that are distinctive for a region of interest, wherein the speckle patterns and / or the one or more other patterns appear in both the first 2D / 3D volume and the second 2D / 3D volume. Further, in some embodiments, the algorithm is further trained on labeled training data for end-to-end prediction of strain from two consecutive 2D / 3D volumes. In some embodiments of the methods, the algorithm is further trained using data comprising a plurality of externally measured strain measurements at a known location in the 2D / 3D volume.

[0027] In some embodiments of the methods, identifying altered strain in the tissue differentiates ablated cardiac tissue from untreated cardiac tissue. In particular, the algorithm generates a pretrained model that predicts strain between two ultrasound volumes, in some embodiments. Further, in some embodiments, the pre-trained model further comprises a segmentation task, wherein the segmentation task differentiates ablated cardiac tissue from untreated tissue. In particular embodiments, the model is trained using 2D / 3D volumes of different time points from a plurality of heartbeat cycles to achieve a base for identifying differences in strain between 2D / 3D volumes, wherein the model is trained before ablation. In other embodiments, the model is trained during ablation, wherein the model comprises one or more of an unsupervised machine learning model and a supervised model trained with labeled data from one or more of elastography data and histology data identifying lesion tissue. In some embodiments of the methods, a difference in strain between 2D / 3D volumes before ablation and during ablation generates a lesion identification, wherein strain differences are predictive of the lesion extent. Further, in some embodiments, the generated lesion identification is output as a volume overlayed on a model of the 2D / 3D volume. Further, the algorithm calculates an ablation depth to generate an ablation depth profile in the tissue over an ablation path, in some embodiments.

[0028] In some embodiments of the methods, the ultrasound data is obtained from a ID or 2D matrix array providing 3D imaging with electronic or mechanic beam steering capabilities. In particular embodiments, the received ultrasound image data comprises full circumferential 3D image data obtained from an intracardiac echography (ICE) catheter, wherein the ultrasound imaging data is real-time ultrasound image data comprising 3D volume data associated with cardiac tissue. Further, in some embodiments, the data comprises one or more of a plurality of time-sequentially acquired B-Images, beamformed RF data, and enveloped data over a full perimeter of a catheter tip, wherein the data corresponds to a cross-section of the tissue during the complete single heartbeat. In some embodiments, the ultrasound image data comprises a complete simultaneously acquired volume around the full perimeter of a catheter tip.

[0029] In some embodiments of the methods, the system is further configured to preprocess the ultrasound image data, wherein preprocess comprises one or more of noise removal, image smoothing, and contrast enhancement.

[0030] In some embodiments of the methods of the invention, analyzing further comprises analyzing data from one or more ultrasound signals at one or more signal stages, wherein the one or more signal stages comprise compounded, envelope, and log-compressed. In some embodiments, analyzing further comprises analyzing data from one or more methods comprising a Nakagami method, and a microstructural envelop statistics method. In some embodiments, analyzing further comprises analyzing data related to one or more of perfusion, stiffness, anisotropy, coherence, specific statistical distributions in tissue, frequency power spectrum of tissue, ultrasound data at various beamforming stages, one or more learned features from raw and / or processed signals, interventional tool tracking data, and breath data.

[0031] Brief Description of the Drawings

[0032] FIG. 1 illustrates examples of the clinical needs during catheter ablation to treat atrial fibrillation.

[0033] FIG. 2 illustrates a high-resolution intracardiac navigation system, i.e. orbital sonography technology, utilized in some embodiments of systems and methods of the invention.

[0034] FIG. 3A and FIG. 3B are diagrammatic illustrations of an ultrasound system according to one embodiment of the invention.

[0035] FIG. 4 is a perspective view of an imaging catheter with which systems of the invention may be coupled.

[0036] FIG. 5 illustrates an overall method for lesion imaging according to some embodiments of the invention.

[0037] FIG. 6 illustrates an output volume of the generated lesion identification. FIG. 7 illustrates a block diagram of a method for analyzing cardiac tissue images according to one embodiment of the invention.

[0038] Detailed Description

[0039] The present invention addresses the limitations of currently utilized approaches for treating arrhythmias using interventional cardiac echography (ICE) ultrasound systems and provides improved systems and methods for lesion detection and analysis. In particular, the invention provides systems and methods for differentiating untreated cardiac tissue versus ablated cardiac tissue during electrophysiology (EP) therapy and / or measuring an ablation depth profde during cardiac ablation to treat arrythmias such as atrial fibrillation (AF). The systems and methods of the invention provide for analyzing cardiac heart function within a single beat to analyze the stress and strain relationship of cardiac tissue to identify areas of altered strain within the tissue. Thus, by identifying areas of altered strain with the tissue, the systems and inventions provide for both identifying a lesion and measuring an ablation depth profile in the tissue over the ablation path.

[0040] The systems and methods of the invention use machine learning approaches in combination with pattern-based tracking methods, a block matching methods, and / or tracking of one or more anatomical regions comprising one or more anatomical landmarks and / or structures in both 2D images and 3D volumes to provide real-time imaging and analysis of the stress and strain of heart movement in all three-dimensional directions. In contrast to conventional approaches, the invention utilizes pattern-based tracking not only in 2-dimensional images but also for 3D volumes.

[0041] The invention recognizes that, despite the extensive utilization of imaging equipment and tools for ablation, as well as systems supporting the identification of electrically active regions within the heart (e.g. electroanatomical mapping systems), conventional imaging systems do not allow a clinical user to understand the complex endocardial and myocardial anatomy live and in sufficient detail to provide a clinician with highly reliable anatomical and physiological feedback before, during, and after an ablation procedure. Accordingly, the systems and methods of the invention address this problem and provide for fast, accurate, and live calculation of a projected lesion depth, which provides lesion extent information to clinicians during the ablation procedure in order to avoid reinterventions. Systems and methods of the invention provide for lesion detection and analysis through analyzing heart tissue dynamics. Thus, the invention provides for intra-procedural feedback to enable patient-specific planning and ablation monitoring compatible with thermal ablation, radio frequency (RF) ablation, pulsed field ablation (PF A), cryoablation , laser ablation, and ultrasound procedures.

[0042] Overview

[0043] The invention provides systems and methods for differentiating ablated tissue from untreated tissue, and for measuring an ablation depth profile in tissue over an ablation path. The systems and methods of the invention provide information about tissue during, for example, electrophysiology (EP) assessment / therapy and / or before, during, and / or after an ablation procedure.

[0044] Conventional approaches, that attempt to integrate myocardial mechanics with three- dimensional imaging, such as 3D speckle-tracking echocardiography, suffer from disadvantages including requiring an adequate temporal resolution to ensure the presence of recognizable natural acoustic markers, and the reliance on patient cooperation for breath holding, which limits the feasibility of these methods in a significant proportion of routine patients. Further, for conventional systems, providing a high frame rate single-beat capability comes at the expense of lower spatial resolution of 3D images.

[0045] The invention recognizes the limitations of conventional approaches to cardiac lesion mapping and solves these problems by utilizing machine learning approaches and pattern-based tracking to differentiate non-ablated tissue from ablated tissue. In particular, the invention applies novel approaches to the analysis of both 2D ultrasound image data as well as 3D volume data. Thus, the invention recognizes that, in a 2D B-Image, using conventional approaches utilizing in-plane speckle movement and its tracking through stress-induced out-of-plane forces, speckles appear and disappear without in-plane movement. Thus, conventional systems routinely include poorly tracked segments in the computation of global strain values and have no clear discrimination between poor tracking quality and poor function in abnormal segments. As such, conventional approaches lack the automated capabilities achieved with the present invention, which provide for fast, accurate tissue analysis.

[0046] The invention solves these problems by acquiring pattern data, such as speckle data, over a 3D volume. This allows for the real-time analysis of stress induced through heart movement which further allows for the analysis of strain through pattern-based tracking in all 3-dimensional directions. Further, the systems and methods of the invention utilize machine learning approaches to generate pre-trained models and labeled models from ablated areas. This supervising generates a strain analysis plus a deeper pattern-based analysis for additional analysis of other aspects of the region of interest, such as anatomical landmarks within the region of interest. Thus, systems and methods of the invention provide an improved level of analysis with identification confidence. Further, the invention may be utilized for cardiac strain measurements in heart cavities, thus eliminating the need for a patient to hold their breathing during the procedure.

[0047] Atrial fibrillation (AF) is a common supraventricular arrhythmia that is characterized by rapid and irregular activation in the atria. Atrial fibrillation, as well as other complex cardiac arrythmias such as atrial flutter or ventricular tachycardia, is defined through irregular heartbeats caused by chaotic electrical signals in atrial or ventricular chambers of the heart. AF is associated with many adverse outcomes, including stroke, dementia, heart failure, increased medical costs, impaired quality of life, and mortality. Paroxysmal AF (PAF) is defined as AF that terminates spontaneously or with intervention within 7 days of onset; persistent AF is defined as continuous AF that is sustained beyond 7 days; and long-standing persistent AF is defined as continuous AF of greater than 12 months’ duration. Silent AF is defined as asymptomatic AF diagnosed by an opportune ECG or rhythm strip. Paroxysmal, persistent, and long-standing persistent AF can be silent.

[0048] The pathophysiology of AF is complex, involving interaction among multiple factors, including triggers, which are responsible for AF initiation; substrate, which is necessary for AF maintenance; and perpetuators, which underlie the progression of the arrhythmia from paroxysmal to the persistent forms. For example, diverse factors contributing to the pathophysiology of AF include oxidative stress, calcium overload, atrial dilatation, microRNAs, inflammation, and myofibroblast activation. The central mechanisms governing AF initiation and perpetuation are poorly understood, which explains in part why treatment of patients with all forms of AF, and particularly long-standing persistent AF, remains suboptimal.

[0049] Catheter ablation of atrial fibrillation is a common yet highly complex procedure for the treatment of arrhythmias. An electrophysiology (EP) analysis is performed before cardiac ablation to identify the area or areas of irregular heart rhythm. Ablative therapy is aimed at either eliminating the trigger initiating AF or modifying the arrhythmogenic substrate. Catheter ablation is a treatment in which energy is applied to cardiac tissue to create scars or lesions for preventing or interrupting the transmission of abnormal electrical signals. In catheter ablation, specific regions within the heart are destroyed resulting in electrical isolation of these regions to prevent a propagation of electrical signals causing the arrhythmia. The most commonly employed ablation strategy consists of electrical isolation of the pulmonary veins by creation of circumferential lesions around the right and the left pulmonary vein. During an ablation procedure applied for treatment of cardiac arrhythmias it is required that thermal or electroporation-based lesion generation lead to electrical conduction blocks in the myocardium. Applied at appropriate locations, this procedure solves arrythmias and the patient’s heart returns to normal sinus triggered rhythm

[0050] As discussed herein, the likelihood of obtaining permanent electrical isolation is related to the quality of ablation energy delivery and lesion formation. There are many factors that play a role in determination of lesion transmurality. For example, with RF energy, common variables that impact transmurality include lesion size, catheter stability, contact force, power output, temperature, and duration of RF output.

[0051] FIG. 1 illustrates examples of the clinical needs during catheter ablation to treat atrial fibrillation. Specifically, ablation performed without direct tissue data feedback may result in gaps in the ablation path and depth, which results in re-entrance of electric signals. The invention addresses these factors and clinical needs. Thus, systems and methods of the invention provide direct feedback on the differentiation of untreated tissue versus ablated cardiac tissue during ablation and / or EP therapy. Further, the systems and methods of the invention provide for calculating an ablation depth profile in tissue over the ablation path to assess for gaps in ablation path and depth, thus providing an ablation depth profile over the ablation path.

[0052] Ultrasound imaging system

[0053] As is generally understood, ultrasound imaging (sonography) uses high-frequency sound waves to view inside the body. Because ultrasound images are captured in real-time, these images can also show movement of the body's internal organs as well as fluid flow (e.g., blood flowing through blood vessels). The imaging device, (i.e. the transducer, probe, or transducer probe) is placed inside a body opening (e.g. endovascular ultrasound, intravascular ultrasound, intracardiac echocardiography). The final quality of the image obtained through ultrasound scanning is limited to the technical specifications of the equipment, the propagation of ultrasonic waves through the tissue analyzed, and the method used to reconstruct the images.

[0054] Systems and methods of the invention may be operably connected with an ultrasound system with certain capabilities for providing image reconstruction and imaging assembly control. As disclosed herein, systems and methods of the invention utilize high-resolution intracardiac navigation systems, for example, systems and methods utilize interventional cardiac echography (ICE) technology with ultrasound to support structural heart and EP therapy. In some embodiments, systems and methods of the invention a cylindrical array, i.e. a matrix of array rows over the complete catheter perimeter for acquisition of ultrasound data.

[0055] FIG. 2 illustrates a high-resolution intracardiac navigation system, i.e. orbital sonography technology, utilized in some embodiments of systems and methods of the invention. Systems and methods of the invention may utilize a system that provides traditional Ultrasound grey-scale images as axial and cross-sections of the 360° volume around the catheter tip, and further visualizes, with 20 Hz temporal resolution, the movement of cardiac structures following the heartbeat, as illustrated in FIG. 1. The 360° view may be visualized as a 3D intensity point cloud and may be turned and moved on the screen along 3 cartesian axes.

[0056] Systems of the invention may be operably connected with an ultrasound system with certain hardware and software for providing image reconstruction and imaging assembly control, for example as described in International PCT Application No. PCT / IB2019 / 000963 (Published as WO 2020 / 044117) to Hennersperger et al., U.S. Application Publication No. US 2022- 0287679A1 to Hennersperger et al., and U.S. Patent No. 11,382,599 to Hennersperger et al., the contents of each which are incorporated by reference herein. The data may be processed using imaging protocols to extract anatomical and functional information, and tissue characteristics as disclosed in more detail herein, and as disclosed in International PCT Application No. PCT / IB2019 / 000963 (Published as WO 2020 / 044117) to Hennersperger et al., U.S. Application Publication No. US 2022-0287679A1 to Hennersperger et al., and U.S. Patent No. 11,382,599 to Hennersperger et al., the contents of each which are incorporated by reference herein.

[0057] The systems include a console configured to be operably associated with one or more devices such as an ultrasound imaging device and to exchange data therewith. The console may comprise a hardware processor coupled to non-transitory, computer-readable memory containing instructions executable by the processor to cause the console to receive data associated with a complex anatomy. For example, the console may be operable to receive a plurality of data, process and combine the data, and reconstruct, based on the processing and combining of the data, an interactive digital model of an imaged anatomy. In some embodiments, the complex anatomy may be cardiac and / or vascular anatomy. In some embodiments, the data comprises both catheter-based ultrasound imaging data and pulse phase data. Thus, the instructions executable by the processor may cause the console to receive data associated with at least one of cardiac and vascular anatomy the data comprising catheter-based ultrasound imaging data, and, in some embodiments, pulse phase data and / or 3D position data.

[0058] The console may be in active communication with a computing system configured to communicate across a network. The computing system or computing device may include one or more processors and memory, as well as an input / output mechanism (i.e., a keyboard, knobs, scroll wheels, or the like) with which a user may interact so as to operate the console, including making adjustments to the ultrasound imaging system, saving images, initialization, continuous 3D image registration, filtering, optimization, 3D fusion, and panoramic image reconstruction.

[0059] The console may generally include one or more processors (e.g., a central processing unit (CPU), a graphics processing unit (GPU), or both) and storage, such as main memory, static memory, or a combination of both, which communicate with each other via a bus or the like. The memory according to embodiments of the invention may include a machine-readable medium on which may be stored one or more sets of instructions (e.g., software) embodying any one or more of the methodologies or functions described herein. The software may also reside, completely or at least partially, within the main memory and / or within the processor during execution thereof by the computer system, the main memory and the processor also constituting machine-readable media. The software may further be transmitted or received over a network via the network interface device.

[0060] During operation, the CPU and / or GPU may control the transmission and receipt of electrical currents, subsequently controlling the emission and receipt of sound waves from the probe. The CPU and / or GPU may also analyze electrical pulses that the probe makes in response to reflected waves coming back and convert this data into images (i.e., ultrasound images) that may then be viewed on a display, which may be an integrated monitor. Such images may also be stored in memory and / or printed via a printer. Systems of the invention are configured to receive three-dimensional (3D) ultrasound image data from an imaging device. In some embodiments, the invention provides for reconstruction of a patient-specific anatomical model for use in minimally invasive procedures in the vasculature. Accordingly, ultrafast ultrasound imaging techniques, such as planewave or diverging wave imaging, may be required to enable imaging within the constraints of the application, particularly for intravascular and / or intracardiac tissue assessment and analysis. These constraints could be posed due to the high temporal update rate as required for effects observed in visualization and tissue characterization, where plane and diverging wave methods enable high imaging rates, commonly also referred to ultrafast imaging approaches. Systems and methods of the invention allow for the direct utilization of all native ultrafast imaging techniques.

[0061] For example, for intracardiac imaging, planewave imaging may refer to an ultrasound imaging modality where, through a flat transmit of all transducer elements (at different angles) from the angular imaging aperture, a plane wave front may traverse the tissue and may be partially scattered back to the transducer. From the received radio frequency (RF) (i.e. channel) data the overall image may be reconstructed at once in parallel by dynamically beamforming the received RF data for each target position.

[0062] Ultrafast ultrasound methods offer imaging at thousands of frames per second limited only by the physical propagation speed of sound waves in tissue, and enable ultrasensitive bloodflow tracking, shear-wave imaging, super-resolution imaging, and other applications. For example, achieving optimal spatial resolution while enabling artifact-free imaging of dynamic cardiac structures requires a careful balance between spatial sampling and volumetric update rate which can only be achieved using ultrafast imaging techniques. Thus, the 3D ultrasound image data received by systems of the invention may be real-time 3D ultrasound data. For example, the data may be full circumferential, three-dimensional (3D) image data.

[0063] While exemplary embodiments describe ultrasound imaging data received from 3D ICE catheters, catheter-based ultrasound imaging as described herein is not limiting. Imaging data may be received from catheter-based ultrasound systems that may include endovascular and / or intravascular ultrasound for in-body applications. This may include systems and imaging data for imaging in large cavities of the heart, coronary and peripheral arteries, as well as other organs such as liver, kidney, and the like. Systems and methods of the invention may be configured to receive information on ablation electrodes, e.g. spline electrodes on basket catheters, or single electrodes on a point by point catheter.

[0064] FIG. 3A and FIG. 3B are diagrammatic illustrations of an exemplary ultrasound system 100 for providing catheter-based ultrasound imaging data specific to a patient 12. The system 100 may include an imaging device 101 equipped with an imaging assembly 104 and a console 106 to which the imaging device 101 is to be connected. The imaging device may be an imaging catheter 102. Accordingly, systems and methods for interactive reconstruction of patientspecific digital model of cardiac and / or vascular anatomy may use a 4D Intracardiac echocardiogram (ICE) system that captures the anatomy of interest.

[0065] As disclosed in detail herein, the imaging device may generally be in the form of an imaging catheter capable of providing imaging and mapping capabilities. Accordingly, such a device may be useful for ultrasound visualization of intravascular and / or intracardiac tissue, which may be particularly useful for catheter-based interventional procedures for assessing the anatomy as well as functional data in relation to a target volume of interest.

[0066] FIG. 4 is a perspective view of an imaging catheter 102 with which systems of the invention may be coupled. The catheter 102 may include a catheter body 108, including proximal and distal portions. The imaging assembly 104 may be provided at the distal portion, for example, generally defining a distal end of an imaging catheter. A handle 110 may be operably associated with the catheter body 108 and allow for an operator (i.e., surgeon or other medical professional) to manipulate and advance the imaging assembly 104 and the catheter body 108 to a desired target site within the patient’s vasculature. The handle 110 may include user-operable inputs for controlling various features and functions of the imaging assembly 104. An interface member 112 may be provided at a distal portion of the catheter body 108. The interface member 112 generally provides a connection between the imaging catheter 102, including the imaging assembly 104 and handle 110, and the console 106 for transmission of signals therebetween. The connection may include at least one of a hardwired and wireless connection, for example.

[0067] As generally understood, the systems and methods of the present invention may be used for ultrasound visualization of tissue of any kind with respect to any kind of procedure in which imaging analysis is used and / or preferred. The imaging device may be useful in carrying out catheter ablation to treat a cardiac condition, such as atrial fibrillation (AF) or the like. For example, in some embodiments, the catheter may include components providing associated capabilities. For example, portions of the catheter may include sensors (e.g., localization and / or tracking sensors) and / or energy delivery elements (e.g., ablation elements).

[0068] Acquisition of ultrasound image data may be from a cylindrical array (matrix of array rows over the complete catheter tip perimeter) replacing the rotating transducer array. However, acquisition of the ultrasound imaging data may be via an imaging catheter that includes a fully rotatable transducer unit comprised of an ultrasound transducer array configured to transmit ultrasound pulses to, and receive echoes of the ultrasound pulses from, surrounding intravascular tissue during a procedure. Systems and methods of the invention may use any real-time 2D or 3D ultrasound system for acquiring ultrasound image data. Such ultrasound transmissions result in a collection of image data which is received by the console and subsequently reconstructed into one or more images providing visualization and characterization of the surrounding intravascular tissue. In particular, the console may utilize image data received from an imaging assembly of the imaging catheter to reconstruct one or more images, including at least 2D and 3D images of the anatomical region of interest (i.e., intravascular and / or intracardiac tissue).

[0069] As discussed in more detail herein, the console may process the received image data utilizing certain imaging protocols and algorithms for reconstructing images and subsequently outputting, via a display, the reconstructed images to an operator depicting visualization of the anatomical region of interest. In addition to providing reconstruction of images based on received image data from the imaging assembly, the console may further provide control over the imaging assembly, including control over the emission of ultrasound pulses therefrom (intensity, frequency, duration, etc.) as well as control over the movement of the ultrasound transducer unit (i.e., controlling rotation, including speed and duration of rotation).

[0070] Systems for analyzing image data associated with cardiac tissue

[0071] Systems of the invention provide automatic, real-time, robust, and accurate intraprocedural feedback to clinicians to enable patient-specific planning and ablation monitoring. In particular, systems of the invention provide tissue analysis using heart tissue dynamics determined via one or more algorithms for analyzing ultrasound image data associated with an anatomical region of interest. Systems of the invention provide for processing the ultrasound image data to reconstruct images of the tissue representing the anatomical region of interest during a complete single heartbeat and analyze a stress to strain relationship of the tissue to identify areas of one or more lesions in the tissue.

[0072] Aspects of the invention provide systems for analyzing image data associated with cardiac tissue. The systems include a hardware processor coupled to non-transitory, computer- readable memory containing instructions executable by the processor. The instructions cause the processor to receive ultrasound image data associated with an anatomical area of interest, and run a tissue analysis algorithm to identify one or more lesions present in tissue at the anatomical area of interest. As disclosed in more detail herein, running the tissue analysis algorithm includes processing the data to generate one or more images of tissue at the anatomical area of interest, wherein the one or more images represent cardiac tissue during a complete single heartbeat; analyzing a stress to strain relationship of the tissue; and identifying altered strain in the tissue and detecting one or more lesions in the tissue based, at least in part, on identified altered strain.

[0073] Received ultrasound image data

[0074] Systems of the invention are configured to receive and process ultrasound imaging data. In some embodiments, the console processor may be configured to receive the 3D image data in real-time or near real-time. The ultrasound imaging data may be intracardiac ultrasound image data. For example, the received ultrasound image data may include full circumferential 3D image data obtained from an intracardiac echography (ICE) catheter, such that the ultrasound imaging data is real-time ultrasound image data comprising 3D volume data associated with cardiac tissue. As such, the systems may use available 3D ICE catheters for acquiring ultra-fast 3D and / or 4D ultrasound image data of a considered cardiac and / or vascular anatomy.

[0075] As discussed in more detail herein, the ultrasound data may be obtained from a ID or 2D matrix array providing 3D imaging with electronic or mechanic beam steering capabilities. For example, the data may include one or more of a plurality of time-sequentially acquired B- Images, beamformed RF data, and enveloped data over a full perimeter of a catheter tip, such that the data corresponds to a cross-section of the tissue during the complete single heartbeat. Thus the ultrasound image data may include a complete simultaneously acquired volume around the full perimeter of a catheter tip.

[0076] Using one or more algorithms, the systems may process the ultrasound image data and / or combine the data with other data to generate a reconstructed digital anatomical model. The systems may process the data to generate one or more 2D or 3D volume images of tissue at the anatomical area of interest such that the 2D / 3D volume images represent cardiac tissue during a complete single heartbeat.

[0077] The systems of the invention may utilize multiple fields-of-view, e.g. cylindrical fields- of-view. For example, the systems may receive 3D ultrasound image data from a field-of-view sufficient to capture the anatomical region of interest and to reconstruct the 3D image in realtime or near real-time. The systems may utilize an orbital field-of-view around the catheter, which may be different than the planar field-of-view of conventional ICE catheters with a matrix transducer.

[0078] Systems of the invention may use cylindrical ultrasound data acquired from a 360° view catheter. Additionally and / or alternatively, the systems may use cylindrical ultrasound data acquired from any 3D volume, e.g. cylindrical fields-of-view acquired with a rotating transducer array or a cylindrical folded transducer matrix. Cylindrical / 360-degree imaging may be applied to provide an extended field-of view (FOV). 3D ICE catheters with forward and with sideways transducer arrays are suitable as long as the field-of-view is sufficiently large to provide for reconstruction of an anatomical region of interest. For example, catheters facing sideways may be used as these catheters also cover the forward direction to some degree due to the opening angle of the ultrasound beam. Any field-of view implementations may be used to provide for, where necessary, imaging of the whole vascular geometry and its surroundings for each volume such that imaging may be 360 degrees around the catheter.

[0079] Systems of the invention may be configured to be used with any ablation device, for example, devices providing thermal ablation, radio frequency (RF) ablation, pulsed-field ablation, cryoablation, laser ablation, and ultrasound procedures. For example, the ablation catheter may use radio frequency (RF) ablation or pulsed-field ablation (PF A). As is known to persons skilled in the art, for RF ablation, tissue destruction occurs from the thermal energy associated with radiofrequency. Pulsed-field ablation uses a train of microsecond duration high amplitude electrical pulses to ablate tissue. For pulsed-field ablation, the electric field is most commonly produced by a high-voltage direct current delivered between two or more electrodes.

[0080] The systems may use an intracardiac echography (ICE) catheter to image the target region where the ablation catheter is positioned. The ICE catheter provides the ultrasound signals to, for example, the console and / or processing unit. In some embodiments, the console and the ICE catheter together may operate as the imaging system.

[0081] Tissue analysis

[0082] The systems of the invention provide for analyzing the received ultrasound data using one or more tissue analysis algorithms to identify one or more lesions present in the tissue at the anatomical area of interest. The tissue analysis algorithm identifies one or more lesions present in tissue at an anatomical area of interest. Running the tissue analysis algorithm includes processing the received ultrasound image data to generate one or more images of tissue at the anatomical region of interest. Generating one or more images of tissue at the anatomical region of interest may include reconstructing an interactive, patient-specific 3D digital anatomical model in real-time and / or near real-time to provide intraoperative feedback to a clinician.

[0083] The one or more images of tissue at the anatomical area of interest may represent cardiac tissue during a complete single heartbeat. Accordingly, the tissue analysis algorithm analyzes a stress and strain relationship of the tissue to identify altered strain in the tissue. By doing so, one or more lesions in the tissue may be detected based at least in part on the identified altered strain.

[0084] The stress on heart tissue may be assumed to be periodically induced through compressing and relieving heart muscle structures (myocardium) such that a resulting local strain within tissue depends on local stiffness. Strain is a result of stress applied on an elastic material e g., a linear elastic material property. In a one-dimensional case, the strain is related to stress and tissue stiffness using the following equation, referred to as Equation (1):

[0085] £ = E X O’ Equation (1)

[0086] Where: s strain

[0087] E : stiffness a: stress

[0088] The invention utilizes the understanding that an analysis of local stress to strain relation leads to an identification of ablated tissue from untreated tissue background. Previous work as detailed in Bunting, et. al., 2018, Cardiac Lesion Mapping In Vivo Using Intracardiac Myocardial Elastography. IEEE Trans Ultrason Ferroelectr Freq Control. 2018 Jan;65(l): 14-20, incorporated by reference herein in its entirety, have shown that lesions induced by radiofrequency ablation have reduced end-systolic strain, and therefore increased stiffness.

[0089] Without being bound by any particular theory, the invention is based on the novel concept that lesions induced by other ablation techniques, such as, but not limited to, pulsed field ablations, also lead to altered tissue strain at the location site. The invention applies an ultrasound pattern-based tracking method, such as a speckle tracking method, or an alternative pattern based tracking method to identify strain differences in tissue. Additionally and / or alternatively, the invention may use methods, e.g. blockmatching and tracking of unique regions (e.g. landmarks, interfaces, etc.) applied to ultrasound image data to identify strain difference in tissue. Thus, over the 3D volume acquired by ultrasound, patterns, such as ultrasound speckle patterns, are visible and may be used as input for strain and stiffness analysis.

[0090] Tissue stiffness may be calculated using shear wave velocimetry to image, with an ICE catheter probe, myocardial stiffness and contractility throughout the cardiac cycle, as is known to persons skilled in the art, for example, as described in Hollender, et, al., 2012, Intracardiac echocardiography measurement of dynamic myocardial stiffness with shear wave velocimetry. Ultrasound Med Biol. 2012 Jul;38(7): 1271 -83, incorporated by reference in its entirety herein. Specifically, an ICE catheter-tip ultrasonic linear array may be used to obtain measures of dynamic myocardial stiffness using shear wave velocimetry.

[0091] Image preprocessing

[0092] In some embodiments, processing the data may generate one or more two-dimensional (2D) or three-dimensional (3D) volume images of tissue at the anatomical area of interest, wherein the one or more 2D / 3D volume images represent cardiac tissue during a complete single heartbeat. Thus, the images may be one or more of two-dimensional (2D) and three-dimensional (3D) volume (2D / 3D volume) images. The systems of the invention may be configured to preprocess the ultrasound image data as described in more detail herein.

[0093] Systems of the invention may be configured to receive data from a 360° catheter. The systems may include different processing steps for the pattern-based tracking and machinelearning analysis depending on how the data is acquired. For example, in some embodiments, the ultrasound system includes a 360° catheter that produces data that includes N time sequentially acquired B-Images over the full perimeter of the catheter tip, i .e. data acquired via a rotating transducer single array with a rotation frequency of f (Hz), which is referred to herein as case (a) for the purposes of describing the pre-processing steps. In some embodiments, the ultrasound system includes a 360° catheter that produces data that includes a complete simultaneously acquired volume around the full perimeter of a catheter tip, i.e. data acquired via a cylindrical array of multiple single array rows over the perimeter. Which is referred to herein as case (b) for the purposes of describing the pre-processing steps.

[0094] The systems may use methods for analyzing image data associated with cardiac tissue that utilize a pattern-based tracking machine learning algorithm. The preprocessing steps for case (a) may include sequentially acquiring several B-Images, i.e., step by step over rotation with a certain degree of space between, showing a cross-section of the tissue to be analyzed during a complete single heartbeat. The data may be beamformed RF data, and / or enveloped data. As with traditional pattern-based tracking, such as speckle tracking, the images then may be preprocessed by removing noise, smoothing the image, and enhancing contrast. Finally, a set of images of one rotation may be used to build up a 3D volume. Coordinate transformations, i.e. cylindrical to cartesian, may be applied.

[0095] The preprocessing steps for case (b) may include acquiring a complete cylindrical ultrasound volume. Certain noise reduction methods may then be applied similar to case (a). For the 3D volume generation, coordinate transformations, i.e. cylindrical to cartesian may be applied.

[0096] The above-referenced 3D volumes for case (a) and case (b) represent a timepoint within a full heartbeat, which is referenced herein as “3Dv(t)”. A complete heartbeat lasts T seconds. The full heartbeat is divided into m time points which gives the equation below, referred to herein as Equation (2):

[0097] ( / ) = tO + (i * ^-) Equation (2)

[0098] Where: tO is the starting point of a heart pulse, which may be identified by ECG. When (i) = m, the next heartbeat cycle starts, (j) represents a heartbeat. For example, using the above notation, 3Dv(t(j = 2, i = 2)) represents the second volume within the second heartbeat. In some embodiments, an end-systolic phase of a heartbeat may be used.

[0099] The 3D volume before ablation is referred to herein as 3DV (before ablation) and the 3D volume after ablation is referred to herein as 3Dva (after ablation) in order to differentiate the 3D volumes before and after ablation. Preprocessing may further include noise removal, image smoothing, and contrast enhancement.

[0100] Calculation of displacement and strain

[0101] Systems of the invention may use a pretrained model to predict strain between two ultrasound volumes. As discussed in more detail herein, the system provides for differentiating ablated cardiac tissue from untreated tissue using, for example, a segmentation algorithm. Pixels belonging to the ablated area may be classified as lesions by the neural networks and pixels in the untreated areas of the volume may be classified as non-ablated tissue.

[0102] In some embodiments, analyzing the stress to strain relationship comprises calculating a stress induced through a displacement of the tissue and a resulting strain of the tissue using one or more of a pattern-based tracking method, a block matching method, and / or tracking of one or more anatomical regions comprising one or more anatomical landmarks and / or structures. As noted, in some embodiments, processing the data generates one or more two-dimensional (2D) or three-dimensional (3D) volume images of tissue at the anatomical area of interest, wherein the one or more 2D / 3D volume images represent cardiac tissue during a complete single heartbeat.

[0103] In some embodiments, the pattern-based tacking method may be a speckle tracking analysis within the 2D / 3D volume, wherein a plurality of ultrasound speckles from the 2D / 3D volume are used as input for calculating the stress and the strain. The complete single heartbeat may be divided into discrete time points such that an individual 2D / 3D volume represents a separate time point within the complete single heartbeat. As discussed in more detail herein, in some embodiments, the algorithm calculates an individual 2D / 3D volume represents a separate time point within the complete single heartbeat.

[0104] The strain estimation model may be used as a pretrained model, with an altered segmentation predictor at the head of a neural network. The segmentation network may then be fine-tuned for the task of differentiation of ablated cardiac tissue versus untreated cardiac tissue.

[0105] Strain is a measure of how much the tissue deforms in response to an applied force or stress. The applied stress comes from the natural compression of the heart during a heartbeat. In particular, the systems of the invention use a pattern-based tracking algorithm, to track, for example, the movement of speckles between 3D volumes. For example, based on the movement of the speckles between time points when stress to the heart tissue is low and high, the algorithm calculates the displacement (movement) of the tissue and the resulting strain (deformation) of the tissue within the 3D volume at each location. Timepoints when tissue stress is low or max are assumed from the contraction pattern. For example, the time point when the stress of the tissue is a minimum or a time point when the stress of the tissue is a maximum is calculated from a contraction pattern identified via an electrocardiogram (ECG) or an image based method.

[0106] In some embodiments, one or more machine learning algorithms may analyze the 2D / 3D volumes for several heartbeats and identify the strongest volume changes. For example, the one or more machine learning algorithms may be trained on a plurality of 2D / 3D volume data comprising a plurality of identified tissue displacements in a plurality of complete single heartbeats, wherein a plurality of strongest displacements in the 2D / 3D volume data are identified. Thus, in some embodiments, the displacement between consecutive 2D / 3D volumes are used to calculate the stress and the strain of the tissue. In some embodiments, the machine learning algorithm(s) may be a neural network-based feature extractor to predict the tissue displacement from a first 2D / 3D volume to a second 2D / 3D volume. Further, the feature extractor may use speckle patterns, and / or one or more other patterns that are distinctive for a region of interest. For example, the speckle patterns and / or the one or more other patterns may appear in both the first 2D / 3D volume and the second 2D / 3D volume.

[0107] As disclosed in more detail herein, in some embodiments, the algorithm may be further trained on labeled training data for end-to-end prediction of strain from two consecutive 2D / 3D volumes. In some embodiments, the algorithm(s) may be further trained using data comprising a plurality of externally measured strain measurements at a known location in the 2D / 3D volume.

[0108] Training strategies for lesion estimation models based on displacement estimations

[0109] For both case (a) and case (b) displacement and strain may be calculated using one or more machine learning algorithms. For example, a neural network based feature extractor may be trained to predict displacements from one image volume to the next image volume. The feature extractor may use speckle patterns, and / or other patterns that are distinctive for a region of interest that appears in both imaging volumes.

[0110] In a second processing step the displacements between consecutive volumes may be used to estimate the strain and stiffness of the tissue. Where real-time capability is not important, existing methods may be used for this estimation. The results of the estimation from existing methods may be used as labels for supervised training of neural networks tasked with the end-to- end prediction of strain from two consecutive imaging volumes. Further, external strain measurements may be used to complement this training approach, e.g. the strain is measured with external devices at known locations in the ultrasound volume, and the strain measurement is used at the extracted location as a ground truth label.

[0111] As discussed in more detail herein, one or more different models may be used.

[0112] The model may be trained before an ablation procedure. For example, one or more machine learning algorithms may be trained to learn patterns in the volumetric data and to predict movement of the data. For example, the one or more machine learning algorithms may be trained to learn patterns of speckle data and to predict speckle motion in the image. The algorithm may be trained using supervised learning techniques, where labeled data is used to train the model, for example via measuring strain through traditional ultrasound elastography.

[0113] The algorithm may be trained using unsupervised learning techniques, where the algorithm learns patterns in the data without labels, e g. learn a mapping from a source image to a destination image, without requiring known displacements from alternative methods.

[0114] The model may be trained with 3D volumes (3Dv) of different time points from many heartbeat cycles to achieve a base for identifying the largest differences in strain between 3D volumes. Training may first take place with 3D volumes before ablation.

[0115] Model training additionally and / or alternatively may take place during ablation. As with training before ablation, the model may use speckle patterns, and / or other patterns distinctive for a region of interest that appears in the imaging volumes. The algorithm may be trained using supervised learning techniques, where labeled data is used to train the model, for example via measuring strain through traditional ultrasound elastography. The supervised model may alternatively and / or additionally be trained using histology data, i.e. identifying lesion tissue. The algorithm may be trained using unsupervised learning techniques, where the algorithm learns patterns in the data without labels. The model may be trained with 3D volumes (3Dv) of different time points from many heartbeat cycles to achieve a base for identifying the largest differences in strain between 3D volumes.

[0116] In non-limiting examples, the below training strategies based on displacement estimators may be used for lesion estimation models.

[0117] As used herein, “Ml” refers to a supervised model to predict displacements, and trained with known displacements obtained from traditional elastography. The model input information includes two 3Dvs. The objective is displacement estimation via minimizing the difference (or similar) between predicted displacements and known displacements. Alternatively, this model may be referred to as “Msl”.

[0118] “M2” refers to an unsupervised model to predict displacements and trained only using ultrasound volumes, without known displacement. The model input is two 3Dvs. The objective is displacement estimation via predicting displacement field between the input 3DVs, deforming the source 3DV by the displacement field, and minimizing difference (or similar) between deformed first 3Dv and second 3Dv. Alternatively, this model may be referred to as” Mui”.

[0119] “Mai” refers to a supervised model (the same as Ml using known displacements) but the recorded data is from a time period in which ablation is actively performed, e.g. displacement estimation at the start and end of ablation. The model input is two 3Dvs at the beginning of ablation, and two 3Dvs at the end of ablation. The objective is displacement estimation at the beginning of ablation and the end of ablation. The model uses losses for displacement estimation at each timepoint and adds a loss that compares the displacements after the ablation with displacements before the ablation

[0120] “Ms2” refers to an additional loss term which compares strain estimates within a lesion area (obtained from histology) with strain estimates outside the lesion area. The model uses supervised information of known displacements, with the task of estimating displacements, computing strain, and segmenting lesions. The model input information is Two 3Dvs at the end of ablation and lesion ground truth. The objective is to minimize difference (or similar) between predicted displacements and known displacements, compute strain, and includes a loss function that measures the difference between strain with the groundtruth lesion compared to the strain outside the groundtruth lesion.

[0121] “Ma3” is the same as M2, but acquired during ablation. Additionally a loss that expects that the strain in the predicted lesion segmentation is different from the strain outside of the predicted lesion region is introduced. The model input information is two 3Dvs at the beginning of ablation and two 3Dvs at the end of ablation. The model uses unsupervised information of displacements and lesions. The objective is displacement estimation via predicting displacement field between the input 3Dvs, deforming the source 3Dv by the displacement field, and minimizing the difference (or similar) between the deformed first 3Dv and the second 3Dv.

[0122] General Model overview: Mdis a model that computes the displacement estimate 512that describe the pixelbased deformation from Image Volume Vlt72. The displacement may be applied to V1to estimate V2. VltV2are two volumes, of the same field of view, where V2is acquired at the time t2= f + At. When computing the displacements At may be small: ~50 to 200 ms.

[0123] MA r2) -> s^2

[0124] General Loss function for displacement estimation overview:

[0125] If displacements <512are known, by, for example, computing them with alternative methods, the training may be formulated as a supervised learning task, directly using the predicted, and known displacement fields. If the displacements are not known, the predicted displacement may be applied on the first Input volume <512(Vi) to predict an estimate of the second volume V2and the loss function uses the original image and the deformed image. This is the unsupervised case.

[0126] Supervised training Unsupervised training

[0127] Adding lesion time information overview:

[0128] When knowledge about the lesion application time is used, two time points may be defined, one at the beginning of the lesion creation, and one at the end of the lesion creation.

[0129] The displacement estimation for both times may be used. A loss term that compares the displacement estimates at the end of the ablation with the displacement estimates at the beginning of the ablation may be added. Adding ground truth lesion information

[0130] When knowledge about the ground truth lesion is available from histology or alternative methods, this information may be used to further supervise the model training. If an image region 6 R3that contains the lesion is known, a loss term on the displacement estimates that checks for a difference in strain, as a derived quantity from displacement, may be added.

[0131] FIG. 5 illustrates an exemplary overall method for lesion imaging according to some embodiments of the invention.

[0132] Differentiation of ablated tissue from untreated tissue

[0133] Identifying altered strain in the tissue may be used to differentiate ablated tissue from untreated cardiac tissue. For example, the difference between 2D / 3D volumes before ablation and during ablation may then be translated into lesion identification using Ma2 with pre-trained models of Ml or M2. Alternatively, Mai may be used with pre-trained models Ml or M2 to generate strain differences as the outcome. The strain differences may be predictive for the lesion extent as they relate to differences in local tissue stiffness.

[0134] Thus, the one or more algorithms may generate a pre-trained model that predicts strain between two ultrasound volumes. Further, the pre-trained model may include a segmentation task, such that the segmentation task differentiates ablated cardiac tissue from untreated tissue. Thus, the task of differentiating ablated cardiac tissue from untreated tissue, may be a segmentation task. The strain estimation model may be used as a pretrained model, with an altered segmentation predictor at the head of the neural network. The segmentation network is then fine-tuned for the task of differentiation of ablated cardiac tissue versus untreated cardiac tissue. For example, pixels belonging to the ablated area may be classified as lesions by the networks, whereas pixels in the untreated part of the volume are classified as non-ablated tissue.

[0135] As disclosed herein, in some embodiments, the model may be trained using 2D / 3D volumes of different time points from a plurality of heartbeat cycles to achieve a base for identifying differences in strain between the 2D / 3D volumes, wherein the model is trained before ablation. Further, when the model may be trained during ablation such that the model may include one or more of an unsupervised machine learning model and a supervised model trained with labeled data from one or more of elastography data and histology data identifying lesion tissue. The differences in strain calculations between 2D / 3D volumes before ablation and during ablation may be used to generate a lesion identification. The strain differences may be used as a predictor of lesion extent. As disclosed in more detail herein, the algorithm may also calculate an ablation depth profile in the tissue over the ablation path.

[0136] The generated lesion identification may be output as a volume overlayed on a model of the ultrasound 2D / 3D volume.

[0137] FIG. 6 illustrates an output volume of the generated lesion identification. The generated lesion identification output volume may be overlayed on a surface rendered model from ultrasound 3D volume, as illustrated in FIG. 6. Thus the generated lesion identification as an output provides real-time data to clinicians during an ablation procedure.

[0138] The invention recognizes the limitations of conventional approaches to cardiac lesion mapping by utilizing machine learning approaches and pattern-based tracking to differentiate non-ablated tissue from ablated tissue. In particular, the invention applies novel approaches to the analysis of both 2D ultrasound image data as well as 3D volume data. Thus, the invention recognizes the problems with conventional approaches, specifically that, in a 2D B-Image, using conventional approaches utilizing in-plane speckle movement and its tracking through stress- induced out-of-plane forces, speckles appear and disappear without in-plane movement. The invention solves this problem by acquiring pattern data, such as speckle data, over a 3D volume. This allows for the analysis of stress induced through heart movement which further allows for the analysis of strain through pattern-based tracking in all 3-dimensional directions. Further, the systems and methods of the invention utilize machine learning approaches to generate pretrained models and labeled models from ablated areas. This supervising generates a mix of strain analysis plus a deeper pattern-based analysis for additional significant factors. Thus, systems and methods of the invention provide an improved level of analysis with identification confidence.

[0139] Methods for analyzing image data associated with cardiac tissue Methods of the invention provide automatic, real-time, robust, and accurate intraprocedural feedback to clinicians to enable patient-specific planning and ablation monitoring. In particular, systems of the invention provide tissue analysis using heart tissue dynamics determined via one or more algorithms for analyzing ultrasound image data associated with an anatomical region of interest. In particular, systems of the invention provide for processing the ultrasound image data to reconstruct images of the tissue representing the anatomical region of interest during a complete single heartbeat and analyze a stress to strain relationship of the tissue to identify areas of one or more lesions in the tissue.

[0140] FIG. 7 illustrates a block diagram of a method 700 for analyzing cardiac tissue images according to one embodiment of the invention.

[0141] Aspects of the invention provide methods for analyzing cardiac tissue images. The methods include receiving 701 ultrasound image data associated with an anatomical area of interest; and running 703 a tissue analysis algorithm to identify 705 one or more lesions present in tissue at the anatomical area of interest. Running the tissue analysis algorithm includes processing 707 the data to generate 709 one or more images of tissue at the anatomical area of interest, wherein the one or more images represent cardiac tissue during a complete single heartbeat; analyzing a stress to strain relationship 711 of the tissue; and identifying altered strain in the tissue 613 and detecting one or more lesions in the tissue based, at least in part, on identified altered strain.

[0142] Received ultrasound image data

[0143] Methods of the invention are configured to receive and process ultrasound imaging data. In some embodiments of the methods, the console processor is configured to receive the 3D image data in real-time or near real-time. The ultrasound imaging data may be intracardiac ultrasound image data. For example, the received ultrasound image data may include full circumferential 3D image data obtained from an intracardiac echography (ICE) catheter, such that the ultrasound imaging data is real-time ultrasound image data comprising 3D volume data associated with cardiac tissue. As such, the methods may use available 3D ICE catheters for acquiring ultra-fast 3D and / or 4D ultrasound image data of a considered cardiac and / or vascular anatomy.

[0144] As discussed in more detail herein, the ultrasound data may be obtained from a ID or 2D matrix array providing 3D imaging with electronic or mechanic beam steering capabilities. For example, the data may include one or more of a plurality of time-sequentially acquired B- Images, beamformed RF data, and enveloped data over a full perimeter of a catheter tip, such that the data corresponds to a cross-section of the tissue during the complete single heartbeat. The ultrasound image data may include a complete simultaneously acquired volume around the full perimeter of a catheter tip.

[0145] Using one or more algorithms, the methods may process the ultrasound image data and / or combine the data with other data to generate a reconstructed digital anatomical model. Using one or more algorithms, the methods may process the data to generate one or more 2D or 3D volume images of tissue at the anatomical area of interest such that the 2D / 3D volume images represent cardiac tissue during a complete single heartbeat.

[0146] The methods of the invention may utilize multiple fields-of-view, e.g. cylindrical fields- of-view. For example, the systems may receive 3D ultrasound image data from a field-of-view sufficient to capture the anatomical region of interest and to reconstruct the 3D image in realtime or near real-time.

[0147] In some embodiments, the methods of the invention may utilize an orbital field-of-view around the catheter, which may be different than the planar field-of-view of conventional ICE catheters with a matrix transducer. Thus, methods of the invention may use cylindrical ultrasound data acquired from a 360° view catheter. Additionally and / or alternatively, the systems may use cylindrical ultrasound data acquired from any 3D volume, e.g. cylindrical fields-of-view acquired with a rotating transducer array or a cylindrical folded transducer matrix. Cylindrical / 360-degree imaging may be applied to provide an extended field-of view (FOV). 3D ICE catheters with forward and with sideways transducer arrays are suitable as long as the field- of-view is sufficiently large to provide for reconstruction of an anatomical region of interest. For example, catheters facing sideways may be used as these catheters also cover the forward direction to some degree due to the opening angle of the ultrasound beam. Any field-of view implementations may be used to provide for, where necessary, imaging of the whole vascular geometry and its surroundings for each volume such that imaging may be 360 degrees around the catheter.

[0148] Methods of the invention may be used with any ablation device. For example, the ablation catheter may use radio frequency (RF) ablation or pulsed-field ablation (PF A). As is known to persons skilled in the art, for RF ablation, tissue destruction occurs from the thermal energy associated with radiofrequency. Pulsed-field ablation uses a train of microsecond duration high amplitude electrical pulses to ablate tissue. For pulsed-field ablation, the electric field is most commonly produced by a high-voltage direct current delivered between two or more electrodes. The methods may be used with an intracardiac echography (ICE) catheter to image the target region where the ablation catheter is positioned. The ICE catheter provides the ultrasound signals to, for example, the console and / or processing unit. In some embodiments, the console and the ICE catheter together may operate as the imaging system.

[0149] Tissue analysis

[0150] The methods of the invention provide for analyzing the received ultrasound data using one or more tissue analysis algorithms to identify one or more lesions present in the tissue at the anatomical area of interest. The tissue analysis algorithm identifies one or more lesion present in tissue at an anatomical area of interest. Running the tissue analysis algorithm includes processing the received ultrasound image data to generate one or more images of tissue at the anatomical region of interest. As disclosed in more detail herein, generating one or more images of tissue at the anatomical region of interest may include reconstructing an interactive, patientspecific 3D digital anatomical model in real-time and / or near real-time to provide intraoperative feedback to a clinician.

[0151] The one or more images of tissue at the anatomical area of interest represent cardiac tissue during a complete single heartbeat. Accordingly, the tissue analysis algorithm analyzes a stress and strain relationship of the tissue to identify altered strain in the tissue. By doing so, one or more lesions in the tissue may be detected based at least in part on the identified altered strain.

[0152] Assuming the stress on heart tissue is periodically induced through compressing and relieving heart muscle structures (myocardium) a resulting local strain within tissue depends on local stiffness. Strain is a result of stress applied on an elastic material e.g., a linear elastic material property. In a one-dimensional case, the strain is related to stress and tissue stiffness using the following equation, referred to as Equation (1):

[0153] £ = E X cr Equation (1)

[0154] Where: s: strain

[0155] E : stiffness a-, stress

[0156] A key assumption is that an analysis of local stress to strain relation leads to an identification of ablated tissue from untreated tissue background. The invention is based on the novel theory that lesions induced by ablation techniques, such as but not limited to pulsed field ablations, also lead to altered tissue strain at the location site. The invention applies an ultrasound pattern-based tracking method, such as a speckle tracking method, or an alternative pattern based tracking method to identify strain differences in tissue. Additionally and / or alternatively, the invention uses methods, e.g. blockmatching and tracking of unique regions (e.g. landmarks, interfaces, etc.), applied to ultrasound image data to identify strain difference in tissue. Thus, over the 3D volume acquired by ultrasound, patterns, such as ultrasound speckle patterns are visible and may be used as input for strain and stiffness analysis.

[0157] Tissue stiffness may be calculated using shear wave velocimetry to image, with an ICE catheter probe, myocardial stiffness and contractility throughout the cardiac cycle, as is known to persons skilled in the art. Specifically, an ICE catheter-tip ultrasonic linear array may be used to obtain measures of dynamic myocardial stiffness using shear wave velocimetry.

[0158] Image preprocessing

[0159] In some embodiments of the methods, processing the data generates one or more two- dimensional (2D) or three-dimensional (3D) volume images of tissue at the anatomical area of interest, wherein the one or more 2D / 3D volume images represent cardiac tissue during a complete single heartbeat. The methods of the invention may be configured to preprocess the ultrasound image data as described in more detail herein.

[0160] As disclosed herein, methods of the invention are configured to receive data from a 360° catheter. The methods include different processing steps for the pattern-based tracking and machine-learning analysis depending on how the data is acquired. For example, in some embodiments, the ultrasound system includes a 360° catheter that produces data that includes N time sequentially acquired B-Images over the full perimeter of the catheter tip, i.e. data acquired via a rotating transducer single array with a rotation frequency of f (Hz), which is referred to herein as case (a) for the purposes of describing the pre-processing steps. In some embodiments, the ultrasound system includes a 360° catheter that produces data that includes a complete simultaneously acquired volume around the full perimeter of a catheter tip, i.e. data acquired via a cylindrical array of multiple single array rows over the perimeter. Which is referred to herein as case (b) for the purposes of describing the pre-processing steps.

[0161] The methods for analyzing image data associated with cardiac tissue utilize a patternbased tracking machine learning algorithm. The preprocessing steps for case (a) may include sequentially acquiring several B-Images, i.e., step by step over rotation with a certain degree of space between, showing a cross-section of the tissue to be analyzed during a complete single heartbeat. The data may be beamformed RF data, and / or enveloped data. As with traditional pattern-based tracking, such as speckle tracking, the images then may be preprocessed by removing noise, smoothing the image, and enhancing contrast. Finally, a set of images of one rotation used to build up a 3D volume. Coordinate transformations, i.e. cylindrical to cartesian, may be applied.

[0162] The preprocessing steps for case (b) may include acquiring a complete cylindrical ultrasound volume. Certain noise reduction methods may then be applied similar to case (a). For the 3D volume generation, coordinate transformations, i.e. cylindrical to cartesian may be applied.

[0163] The above-referenced 3D volumes for case (a) and case (b) represent a timepoint within a full heartbeat, which is referenced herein as “3Dv(t)”. A complete heartbeat lasts T seconds. The full heartbeat is divided into m time points which gives the equation below, referred to herein as Equation (2): Equation (2)

[0164] Where: tO is the starting point of a heart pulse, which may be identified by ECG. When (i) = m, the next heartbeat cycle starts. ( / ) represents a heartbeat. For example, using the above notation, 3Dv(t(j = 2, i = 2)) represents the second volume within the second heartbeat. In some embodiments, an end-systolic phase of a heartbeat may be used.

[0165] The 3D volume before ablation is referred to herein as 3DV (before ablation) and the 3D volume after ablation is referred to herein as 3Dva (after ablation) in order to differentiate the 3D volumes before and after ablation. As disclosed herein the images may be one or more of two-dimensional (2D) and three- dimensional (3D) volume (2D / 3D volume) images. Preprocessing may further include noise removal, image smoothing, and contrast enhancement.

[0166] Calculation of displacement and strain

[0167] Methods of the invention may use a pretrained model to predict strain between two ultrasound volumes. As discussed in more detail herein, the methods provide for differentiating ablated cardiac tissue from untreated tissue using, for example, a segmentation algorithm. Pixels belonging to the ablated area may be classified as lesions by the neural networks and pixels in the untreated areas of the volume may be classified as non-ablated tissue.

[0168] In some embodiments of the methods, analyzing the stress to strain relationship comprises calculating a stress induced through a displacement of the tissue and a resulting strain of the tissue using one or more of a pattern-based tracking method, a block matching method, and / or tracking of one or more anatomical regions comprising one or more anatomical landmarks and / or structures. As noted, in some embodiments, processing the data generates one or more two-dimensional (2D) or three-dimensional (3D) volume images of tissue at the anatomical area of interest, wherein the one or more 2D / 3D volume images represent cardiac tissue during a complete single heartbeat.

[0169] In some embodiments, the pattern-based tacking method may be a speckle tracking analysis within the 2D / 3D volume, wherein a plurality of ultrasound speckles from the 2D / 3D volume are used as input for calculating the stress and the strain. The complete single heartbeat may be divided into discrete time points such that an individual 2D / 3D volume represents a separate time point within the complete single heartbeat. In some embodiments, the algorithm may calculate an individual 2D / 3D volume representing a separate time point within the complete single heartbeat.

[0170] The strain estimation model may be used as a pretrained model, with an altered segmentation predictor at the head of a neural network. The segmentation network may then be fine-tuned for the task of differentiation of ablated cardiac tissue versus untreated cardiac tissue.

[0171] Strain is a measure of how much the tissue deforms in response to an applied force or stress. The applied stress comes from the natural compression of the heart during a heartbeat. In particular, the systems of the invention use a pattern-based tracking algorithm, to track, for example, the movement of speckles between 3D volumes. For example, based on the movement of the speckles between time points when stress to the heart tissue is low and high, the algorithm calculates the displacement (movement) of the tissue and the resulting strain (deformation) of the tissue within the 3D volume at each location. Timepoints when tissue stress is low or max are assumed from the contraction pattern. For example, the time point when the stress of the tissue is a minimum or a time point when the stress of the tissue is a maximum is calculated from a contraction pattern identified via an electrocardiogram (ECG) or an image based method.

[0172] In some embodiments, one or more machine learning algorithms may be used to analyze the 2D / 3D volumes for several heartbeats and identify the strongest volume changes. For example, the one or more machine learning algorithms may be trained on a plurality of 2D / 3D volume data comprising a plurality of identified tissue displacements in a plurality of complete single heartbeats, wherein a plurality of strongest displacements in the 2D / 3D volume data are identified. Thus, in some embodiments, the displacement between consecutive 2D / 3D volumes are used to calculate the stress and the strain of the tissue. In some embodiments, the machine learning algorithm(s) may be a neural network-based feature extractor to predict the tissue displacement from a first 2D / 3D volume to a second 2D / 3D volume. Further, the feature extractor may use speckle patterns, and / or one or more other patterns that are distinctive for a region of interest. For example, the speckle patterns and / or the one or more other patterns may appear in both the first 2D / 3D volume and the second 2D / 3D volume.

[0173] As disclosed in more detail herein, in some embodiments, the algorithm may be further trained on labeled training data for end-to-end prediction of strain from two consecutive 2D / 3D volumes. In some embodiments, the algorithm(s) may be further trained using data comprising a plurality of externally measured strain measurements at a known location in the 2D / 3D volume.

[0174] Training strategies for lesion estimation models based on displacement estimations

[0175] For both case (a) and case (b) displacement and strain may be calculated using one or more machine learning algorithms. For example, a neural network based feature extractor may be trained to predict displacements from one image volume to the next image volume. The feature extractor may use speckle patterns, and / or other patterns that are distinctive for a region of interest that appears in both imaging volumes.

[0176] In a second processing step the displacements between consecutive volumes may be used to estimate the strain and stiffness of the tissue. Where real-time capability is not important, existing methods may be used for this estimation. The results of the estimation from existing methods may be used as labels for supervised training of neural networks tasked with the end-to- end prediction of strain from two consecutive imaging volumes.

[0177] Further, external strain measurements may be used to complement this training approach, e.g. the strain is measured with external devices at known locations in the ultrasound volume, and the strain measurement is used at the extracted location as a ground truth label.

[0178] As discussed in more detail herein, one or more different models may be used.

[0179] The model may be trained before an ablation procedure. For example, one or more machine learning algorithms may be trained to learn patterns in the volumetric data and to predict movement of the data. For example, the one or more machine learning algorithms may be trained to learn patterns of speckle data and to predict speckle motion in the image. The algorithm may be trained using supervised learning techniques, where labeled data is used to train the model, for example via measuring strain through traditional ultrasound elastography.

[0180] The algorithm may be trained using unsupervised learning techniques, where the algorithm learns patterns in the data without labels, e.g. learn a mapping from a source image to a destination image, without requiring known displacements from alternative methods.

[0181] The model may be trained with 3D volumes (3Dv) of different time points from many heartbeat cycles to achieve a base for identifying the largest differences in strain between 3D volumes. Training may first take place with 3D volumes before ablation.

[0182] Model training additionally and / or alternatively may take place during ablation. As with training before ablation, the model may use speckle patterns, and / or other patterns distinctive for a region of interest that appears in the imaging volumes. The algorithm may be trained using supervised learning techniques, where labeled data is used to train the model, for example via measuring strain through traditional ultrasound elastography. The supervised model may alternatively and / or additionally be trained using histology data, i.e. identifying lesion tissue. The algorithm may be trained using unsupervised learning techniques, where the algorithm learns patterns in the data without labels. The model may be trained with 3D volumes (3Dv) of different time points from many heartbeat cycles to achieve a base for identifying the largest differences in strain between 3D volumes.

[0183] In non-limiting examples, the below training strategies based on displacement estimators may be used for lesion estimation models. As used herein, “Ml” refers to a supervised model to predict displacements, and trained with known displacements obtained from traditional elastography. The model input information includes two 3Dvs. The objective is displacement estimation via minimizing the difference (or similar) between predicted displacements and known displacements. Alternatively, this model may be referred to as “Msl”.

[0184] “M2” refers to an unsupervised model to predict displacements and trained only using ultrasound volumes, without known displacement. The model input is two 3Dvs. The objective is displacement estimation via predicting displacement field between the input 3DVs, deforming the source 3DV by the displacement field, and minimizing difference (or similar) between deformed first 3Dv and second 3Dv. Alternatively, this model may be referred to as” Mui”.

[0185] “Mai” refers to a supervised model (the same as Ml using known displacements) but the recorded data is from a time period in which ablation is actively performed, e g. displacement estimation at the start and end of ablation. The model input is two 3Dvs at the beginning of ablation, and two 3Dvs at the end of ablation. The objective is displacement estimation at the beginning of ablation and the end of ablation. The model uses losses for displacement estimation at each timepoint and adds a loss that compares the displacements after the ablation with displacements before the ablation

[0186] “Ms2” refers to an additional loss term which compares strain estimates within a lesion area (obtained from histology) with strain estimates outside the lesion area. The model uses supervised information of known displacements, with the task of estimating displacements, computing strain, and segmenting lesions. The model input information is Two 3Dvs at the end of ablation and lesion ground truth. The objective is to minimize difference (or similar) between predicted displacements and known displacements, compute strain, and includes a loss function that measures the difference between strain with the groundtruth lesion compared to the strain outside the groundtruth lesion.

[0187] “Ma3” is the same as M2, but acquired during ablation. Additionally a loss that expects that the strain in the predicted lesion segmentation is different from the strain outside of the predicted lesion region is introduced. The model input information is two 3Dvs at the beginning of ablation and two 3Dvs at the end of ablation. The model uses unsupervised information of displacements and lesions. The objective is displacement estimation via predicting displacement field between the input 3Dvs, deforming the source 3Dv by the displacement field, and minimizing the difference (or similar) between the deformed first 3Dv and the second 3Dv.

[0188] General Model overview:

[0189] Mdis a model that computes the displacement estimate <512that describe the pixelbased deformation from Image Volume VltV2. The displacement may be applied to V1to estimate V2. V , V are two volumes, of the same field of view, where 72is acquired at the time t2— tr+ At. When computing the displacements At may be small: ~50 to 200 ms.

[0190] General Loss function for displacement estimation overview:

[0191] If displacements <512are known, by, for example, computing displacements with alternative methods, the training may be formulated as a supervised learning task, directly using the predicted, and known displacement fields. If the displacements are not known, the predicted displacement may be applied on the first Input volume <5I2(VI) to predict an estimate of the second volume V2and the loss function uses the original image and the deformed image. This is the unsupervised case.

[0192] Supervised training unsupervised training

[0193] Adding lesion time information overview:

[0194] When knowledge about the lesion application time is used, two time points may be defined, one at the beginning of the lesion creation, and one at the end of the lesion creation. tt= start of lesion creation, t2= end of lesion creation The displacement estimation for both times may be used. A loss term that compares the displacement estimates at the end of the ablation with the displacement estimates at the beginning of the ablation may be added.

[0195] Adding ground truth lesion information

[0196] When knowledge about the ground truth lesion is available from histology or alternative methods, this information may be used to further supervise the model training. If an image region G R3that contains the lesion is known, a loss term on the displacement estimates that checks for a difference in strain, as a derived quantity from displacement, may be added.

[0197] Again, referring back to FIG. 5, an overall method for legion imaging according to some embodiments of the invention is illustrated therein.

[0198] Differentiation of ablated tissue from untreated tissue

[0199] Identifying altered strain in the tissue may be used to differentiate ablated tissue from untreated cardiac tissue. Thus, identifying altered strain in the tissue may be used to detect one or more lesion in the tissue, based in part on the identified altered strain. For example, the difference between 2D / 3D volumes before ablation and during ablation may then be translated into lesion identification using Ma2 with pre-trained models of Ml or M2. Alternatively, Mai may be used with pre-trained models Ml or M2 to generate strain differences as the outcome. The strain differences may be predictive for the lesion extent as they relate to differences in local tissue stiffness.

[0200] Thus, the one or more algorithms may generate a pre-trained model that predicts strain between two ultrasound volumes. Further, the pre-trained model may include a segmentation task, such that the segmentation task differentiates ablated cardiac tissue from untreated tissue. Thus, the task of differentiating ablated cardiac tissue from untreated tissue, may be a segmentation task. The strain estimation model may be used as a pretrained model, with an altered segmentation predictor at the head of the neural network. The segmentation network is then fine-tuned for the task of differentiation of ablated cardiac tissue versus untreated cardiac tissue. For example, pixels belonging to the ablated area may be classified as lesions by the networks, whereas pixels in the untreated part of the volume are classified as non-ablated tissue.

[0201] As disclosed herein, in some embodiments of the methods, the model may be trained using 2D / 3D volumes of different time points from a plurality of heartbeat cycles to achieve a base for identifying differences in strain between the 2D / 3D volumes, wherein the model is trained before ablation. Further, when the model may be trained during ablation such that the model may include one or more of an unsupervised machine learning model and a supervised model trained with labeled data from one or more of elastography data and histology data identifying lesion tissue. The differences in strain calculations between 2D / 3D volumes before ablation and during ablation may be used to generate a lesion identification. The strain differences may be used as a predictor of lesion extent. As disclosed in more detail herein, the algorithm may also calculate an ablation depth profile in the tissue over the ablation path.

[0202] The generated lesion identification may be output as a volume overlayed on a model of the ultrasound 2D / 3D volume. The generated lesion identification as an output may provide realtime, near real-time, and / or interactive data to clinicians during an ablation procedure.

[0203] Reconstruction of patient-specific 3D digital anatomical model

[0204] As disclosed herein, processing the data generates one or more two-dimensional (2D) or three-dimensional (3D) volume images of tissue at the anatomical area of interest, such that the one or more 2D / 3D volume images represent cardiac tissue during a complete single heartbeat. The invention uses a pattern-based tracking algorithm, to track, for example, the movement of speckles between 3D volumes, a block matching tracking method, and / or tracking of one or more anatomical regions comprising one or more anatomical landmarks and / or structures. Thus, systems and methods of the invention may generate an interactive and / or real-time patientspecific 3D digital anatomical model of the target region of interest using one or more algorithms to achieve the 2D / 3D volume images.

[0205] For example, in some embodiments, the systems and methods of the invention are configured to run, via the console one or more imaging algorithms configured to analyze the 2D / 3D ultrasound image data and identify one or more anatomical structures of interest. Systems and methods of the invention generate a reliable and anatomically correct real-time or near real-time patient-specific 3D digital anatomical model. The systems and methods of the invention may receive, process, and combine real-time catheter-based ultrasound imaging data with other data to reconstruct a representation of the digital anatomy.

[0206] For example, systems and methods of the invention may adapt and utilize one or more image segmentation algorithms for generating a patient-specific digital anatomical model. Image segmentation is the process of dividing an image into multiple meaningful and homogeneous regions or objects based on their inherent characteristics, such as color, texture, shape, or brightness. Each pixel may be labeled, and all pixels belonging to the same category may have a common label assigned to them. Segmentation may be achieved via instance image segmentation in which each object in an image is detected and segmented, via one or more algorithms to separate overlapping objects. Segmentation may be achieved via semantic segmentation in which one or more algorithms are used to label each pixel. Segmentation may be achieved via panoptic segmentation in which one or more machine learning algorithms are used to label each pixel with a class label and to identify each object instance in the image to provide for detection and interaction of the object within the environment. CV-based techniques may be used to combine multiple image data into an anatomical representation that combines the multiple image data into a large anatomical representation. Segmentation may be performed by a vision algorithm such as thresholding, connected component analysis, or a neural network based segmentation. Segmentation of the full panoramic volumes may be performed, for example, via a deep learning algorithm.

[0207] In some embodiments, the segmentation algorithm may identify the anatomical surrounding and the ablation catheter in the 3D image data, such that a region of interest with anatomical context may be provided. The region of interest may define an area where tissue wall thickness will be reconstructed. In some embodiments, the one or more imaging algorithms are configured to calculate the position and orientation of the ablation catheter relative to a targeted tissue within a heart anatomy such that an exact position of the ablation catheter and the electrode tip relative to the targeted tissue may be identified. For example, systems and methods of the invention may generate a patient-specific 3D digital anatomical model based on 4D image data acquired along an ICE catheter’s trajectory over time, and provide for navigation of the ICE catheter to the target region of interest. In some embodiments, the systems and methods utilize medical image registration, computer vision (CV)-based approaches, and / or simultaneous localization and mapping (SLAM).

[0208] Simultaneous localization and mapping (SLAM) is a computational method that constructs or updates a map of an unknown environment while simultaneously keeping track of an agent’s location within it. In some embodiments, systems and methods of the invention adapt and utilize one or more algorithms for SLAM applications to ultrasound image data acquired to generate the patient-specific anatomical model as well as to navigate the catheter. For example, SLAM processing techniques may be adapted for use in conjunction with the catheter utilizing various sensors to incrementally build the map of a patient anatomical environment and simultaneously determine the location of the catheter within the map. Thus, the systems and methods of the invention may use adapted SLAM techniques to combine multiple image data into an anatomical representation that combines the multiple image data into a large anatomical representation.

[0209] CV-based image reconstruction approaches utilized by systems of the invention may include approaches based on deep neural networks (DNNs) such as autoencoders (AEs), convolutional neural networks (CNNs), and generative adversarial networks (GANs). The computer vision approaches utilized by systems of the invention aim to detect, interpret and reconstruct data in a way that mimics the intricacy of the human visual system thus providing for intuitive navigation to precisely target an anatomical region.

[0210] Reconstruction of an interactive digital model of an imaged anatomy and / or region of interest may include initialization, continuous 3D image registration, filtering, optimization, 3D fusion, and panoramic image reconstruction. The systems and methods may use multimodal image registration using one or more algorithms to correlate morphologic and / or functional features between images. The systems and methods may filter the 3D ultrasound images, subsequent to initialization, along the ICE catheter’s trajectory. For example, filtering may include using pulse phase-gating. The pulse-phase gating may be ECG-gating in some embodiments. Continuous registration may include registering 3D ultrasound images of the same cardiac phase against the respective previous 3D ultrasound image or a current fused 3D image. Image registration is the process of aligning multiple data, i.e. images, volumes, or surfaces to a patient coordinate system. Systems and methods of the invention may utilize one or more registration algorithms to, for example, combine images of the patient and / or data from different modalities and to align temporal sequences of images to generate the interactive and / or real-time patient-specific digital anatomical model. The one or more algorithms find an optimal spatial transformation that best aligns the underlying anatomical structures for reconstruction of the digital anatomical model. Thus, the systems and methods of the invention may use image registration techniques to combine multiple image data into an anatomical representation that combines the multiple image data into a large anatomical representation.

[0211] Systems and methods of the invention may receive 3D position and orientation data of the catheter, and process and combine this data with the ultrasound imaging data and / or other data, such as pulse-phase data, for reconstruction of the digital anatomical model. It is noted however, that in some embodiments, reconstruction of the digital anatomical model is possible without 3D position data. The 3D position and orientation data may be referred to as 3D pose data. 3D pose data may be 6 degree-of-freedom tracking including position and orientation. The 3D pose data may comprise a position and orientation in 3D space. Tracking data may be used by the console to map the ultrasound image data to the patient coordinate system. Accordingly, the systems and methods of the invention may provide for the real-time 3D localization and tracking of the catheter, for example an ICE catheter, and interventional tools within the patient coordinate system.

[0212] The one or more algorithms may include identifying the position / orientation and / or location of the ablation catheter and its relation to heart anatomy. Calculating a position and / or orientation of the ablation catheter may include receiving data from one or more position sensors. For example, the ablation catheter position and / or orientation may be calculated using positional sensing.

[0213] The 3D pose data may be obtained through electromagnetic (EM) tracking, impedance tracking, image-based tracking, fiber-optic shape sensing, and / or a combination of these modalities. In some embodiments, EM tracking data is used. Electromagnetic tracking generates a defined EM field in which EM micro sensors are tracked. 6 degree-of-freedom tracking information (spatial position and orientation) may be acquired, for example, by embedding micro sensors into rigid or flexible instruments, where they serve as localization points for the instrument in space. The micro sensors may be embedded, for example, in a coil in the catheter tip. This allows for tracking the catheter tip position inside an electromagnetic field. The EM field generator emits a low intensity, varying EM field that establishes a measurement volume. Small currents are induced inside the sensors when they enter the EM field. The currents are relayed to the sensor interface unit where they are amplified and digitized as signals. The signals are transmitted to the console which calculates each sensor’s position and orientation as a transformation.

[0214] In some embodiments EM tracking of interventional tools is also incorporated. Thus, in some embodiments, the 3D position data comprises a spatial position and an orientation of a catheter and / or an interventional tool. In some embodiments, the 3D pose data is obtained through optical fiber shape sensing. For example, in some embodiments the fiber optic shape sensing comprises a fiber Bragg grating (FBG) sensor. For example, low reflectance fiber Bragg grating (FBG) strain sensors may be positioned in a multi-core fiber within the catheter to determine how a point along the fiber is positioned in space. By sensing the relative change of FBGs in each of three or more fiber cores, the three-dimensional position may be determined.

[0215] The one or more algorithms may include subsequent optimization steps. Subsequent optimization steps increase the accuracy of the localization of each image in the patient’s coordinate system. The optimization steps also ensure robustness of the system. For example, where 3D pose data comprises EM tracking data, optimization ensures robustness of the systems against missing EM tracking data, for instance where the ICE catheter leaves the EM tracking field. To provide for improved robustness of registration in the case of inaccurate EM data used for initialization, or in the case of limited image features (e.g. when the ICE catheter is in the inferior vena cava (IVC) next to the lung), an additional tracking scheme may be used. The tracking scheme may use the EM tracking data and the previous registration results to predict the next 3D position. The predicted next 3D position may then be used for initialization of the continuous registration. To further provide for improved registration in regions with few features, prior knowledge of the imaged anatomy may be applied.

[0216] The systems and methods may combine the received data and further continuously register subsequent 3D ultrasound images within the physical patient’s coordinate system, such that the continuously registered subsequent 3D ultrasound images may be partly overlapping. The continuously registered and partly overlapping subsequent 3D ultrasound images may then be fused into a large 3D image representing a patient-specific panoramic reconstruction of the anatomy. This 3D panoramic reconstruction may be updated sequentially over time as the catheter is moved through the anatomical region or regions of interest, such that the patientspecific digital 3D anatomical model may be continuously updated with subsequent registration and segmentation results.

[0217] The digital anatomical model may encapsulate a representation of the anatomy in a spatial topology for live visualization of one or more anatomical regions of interest. The spatial topology may include a topological map, such that each point in the topological map represents tissue and / or one or more specific anatomical properties. Thus, ultrasound imaging data from one or more directions and / or one or more views may be combined to provide the representation of the anatomy. This representation may be a panoramic image reconstruction (i.e. an intensity volume), a (segmented) surface model, and / or a mesh. The digital anatomy may also be a more advanced representation of the anatomy, where for each point in the topological map, a representation of tissue or specific anatomical properties are encapsulated, such that different views or information from different directions may be combined for a complete representation. Thus, by reconstructing an interactive digital anatomical model, based on the processing and combining of the received data, the systems of the invention provide a digital anatomical model similar to computed tomography (CT), magnetic resonance imaging (MRI), or Ultrasound Tomographic reconstructions. The digital model encapsulates a representation of the anatomy in a spatial topology for live visualization of one or more anatomical regions of interest for fast and accurate calculation of a tissue wall thickness and / or a distance of the electrode tip from the targeted tissue may be calculated and a contact point of the electrode with the targeted tissue may be determined based on said calculated distance.

[0218] The systems and methods may generate one or more 3D anatomical landmarks. To aid the navigation of the ICE catheter and interventional tools, a detection and localization step may be added to automatically find prominent anatomical landmarks. These landmarks may be visualized in 3D along with the model. The systems and methods may provide for tracking of interventional tools within the 3D model. The console may be further configured to detect, localize, and segment in the 3D ultrasound images, relevant interventional tools. The segmented tools may visualized along with the 3D digital anatomical model. In addition to physical tracking and navigation, the invention provides for contact assessment between the tool tip and cardiac or vascular wall.

[0219] In some embodiments, the model may be generated once when all of the 3D ultrasound image data is available. After model generation, navigation of the ICE catheter as well as tracking and navigation of interventional tools is possible. The anatomical context generated with the model enables safer navigation of catheter tools.

[0220] Thus, as disclosed herein, systems and methods of the invention may receive 3D position and orientation data of the catheter, and process and combine this data with the ultrasound imaging data and / or other data for reconstruction of a patient-specific digital anatomical model. The position and orientation of the catheter in combination with imaging data from ablation catheters (e.g. spline electrodes on basket catheters, or single electrodes on a point by point catheter in some embodiments) is used to reconstruct the tissue wall thickness and to provide feedback to a clinician before and during an ablation procedure.

[0221] As disclosed herein, the systems and methods of the invention may combine one or more modalities for differentiating untreated cardiac tissue from ablated cardiac tissue during electrophysiology (EP) therapy and / or during cardiac ablation to characterize heart ablation. In some embodiments strain measurement data plus data from one or more other methods, e.g. Nakagami or Rayleigh distribution parameters, may be used. In non-limiting examples, one or more data may be used for example, envelop statistics data, such as Nakagami, may be used in static scenarios or masked scenarios such as a wall mask, and data may be included to further define a lesion width or depth for heart ablation characterization. As is known to persons skilled in the art the Nakagami method uses a statistical distribution of backscattered signals to model, analyze, and assess tissue characteristics. Further, data from micro-structural methods, such as elastography, may be included, for example to identify regions without tracking. The envelope statistics may refer to the statistical distribution of ultrasound backscattered envelope signals, wherein the information contained within the envelope of the ultrasound signal is used to characterize tissue microstructure

[0222] As disclosed herein, the systems and methods accomplish lesion characterization by utilizing data related to one or more of the extraction of perfusion, stiffness, strain, anisotropy, coherence, specific statistical distributions in tissue (Nakagami), spectral parameters of tissue (frequency power spectrum) and other parameters. Further, the systems and methods of the invention may utilize analysis methods that incorporate data such as ultrasound data at various beamforming stages, learned features from raw and / or processed signals, deep-learning-based methods incorporating 2D, 3D, and 4D deep learning, compounded, envelope, and / or log- compressed signal beamforming stage, full 4D temporal evaluation for dimensionality, tracking integration, i.e. integrated catheter and breathing tracking, position / width / depth data for lesion output. The captured data may be processed using the disclosed protocols to extract anatomical and functional information, and tissue characteristics.

[0223] As disclosed herein, the systems and methods of the invention provide for improved pullback sequence visualization, i.e. during slowly retracting the imaging catheter from the superior vena cava (SVC) through the right atrium (RA) to the inferior vena cava (IVC) to capture the full cardiac anatomy. The systems and methods of the invention provide for correcting tracking error and breathing motion during a pullback sequence. The systems and methods of the invention recognize that full correction of the breathing motion is likely not possible with rigid registration alone. Thus, in some embodiments, the systems and methods utilize a rigid pairwise registration using tracking data as an initial estimate. The systems and methods may use one or more modalities for preprocessing for registration such as, for example, ECG-gating to select corresponding frames, iMAP2 beamforming3 to minimize the presence of aberrations, reducing speckle pattern using the speckle2speckle filter, and utilizing a similarity metric, i.e. mutual information.

[0224] As disclosed herein, the systems and methods of the invention may use one or more data sources and modalities for tissue characterization. The systems and methods of the invention may use ultrasound imaging modalities at various beam-forming stages. The systems and methods of the invention may use a measurement type that includes learned features from raw / processed signals. The systems and methods of the invention may include 2D / 3D / 4D deep learning in providing live visualization of complex anatomy in sufficient detail for intuitive navigation to precise locations within the anatomy. Systems and methods of the invention may use imaging data and various other data combined with image segmentation and registrationbased or mapping-based techniques as well as 2D / 3D / 4D deep learning to provide for the reconstruction of a real-time or near real-time interactive patient-specific digital anatomical model. For example, in some embodiments, the systems and methods of the invention combine custom feature extraction with deep learning for lesion characterization. Systems and methods of the invention may utilize a compounded, envelope, and / or log-compressed signal stage. Systems of the invention may use full temporal evolution for increased dimensionality. The systems of the invention may also integrate catheter and breathing tracking data, and incorporate tracking data in a 4D context. The systems of the invention provide for position, width, and depth for displaying lesion (lesion output). Thus, the systems and methods of the invention provide for outputting quantitative lesion dimensions rather than a simple classification as output.

[0225] As disclosed herein, the systems and methods of the invention provide for combining strain data with one or more other data sources for analyzing cardiac heart function within a single beat to determine the stress and strain relationship of cardiac tissue and to identify areas of altered strain within the tissue. Thus, the invention provides for both identifying a lesion and for measuring an ablation depth profile in the tissue over the ablation path. As disclosed herein, strain data may be estimated using one or more strain estimation methods. In non-limiting examples, a normalized cross-correlation (1-D cross-correlation with Lagrangian tracking) may be used to estimate inter-frame displacements through the myocardium, and / or select systole may be chosen manually based on the ECG displacements allowing for inter-frame displacements to be accumulated through systole and converted to systolic strain using a leastsquares estimator.

[0226] Any of the operations described herein may be implemented in a system that includes one or more storage mediums having stored thereon, individually or in combination, instructions that when executed by one or more processors perform the methods. Here, the processor may include, for example, a server CPU, a mobile device CPU, and / or other programmable circuitry.

[0227] Also, it is intended that operations described herein may be distributed across a plurality of physical devices, such as processing structures at more than one different physical location. The storage medium may include any type of tangible medium, for example, any type of disk including hard disks, floppy disks, optical disks, compact disk read-only memories (CD-ROMs), compact disk rewritables (CD-RWs), and magneto-optical disks, semiconductor devices such as read-only memories (ROMs), random access memories (RAMs) such as dynamic and static RAMs, erasable programmable read-only memories (EPROMs), electrically erasable programmable read-only memories (EEPROMs), flash memories, Solid State Disks (SSDs), magnetic or optical cards, or any type of media suitable for storing electronic instructions. Other embodiments may be implemented as software modules executed by a programmable control device. The storage medium may be non-transitory.

[0228] As described herein, various embodiments may be implemented using hardware elements, software elements, or any combination thereof. Examples of hardware elements may include processors, microprocessors, circuits, circuit elements (e.g., transistors, resistors, capacitors, inductors, and so forth), integrated circuits, application specific integrated circuits (ASIC), programmable logic devices (PLD), digital signal processors (DSP), field programmable gate array (FPGA), logic gates, registers, semiconductor device, chips, microchips, chip sets, and so forth.

[0229] Reference throughout this specification to “one embodiment” or “an embodiment” means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Thus, appearances of the phrases “in one embodiment” or “in an embodiment” in various places throughout this specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0230] The term "non-transitory" is to be understood to remove only propagating transitory signals per se from the claim scope and does not relinquish rights to all standard computer- readable media that are not only propagating transitory signals per se. Stated another way, the meaning of the term "non-transitory computer-readable medium" and "non-transitory computer- readable storage medium" should be construed to exclude only those types of transitory computer-readable media which were found in In Re Nuijten to fall outside the scope of patentable subject matter under 35 U.S.C. § 101.

[0231] The terms and expressions which have been employed herein are used as terms of description and not of limitation, and there is no intention, in the use of such terms and expressions, of excluding any equivalents of the features shown and described (or portions thereof), and it is recognized that various modifications are possible within the scope of the claims. Accordingly, the claims are intended to cover all such equivalents.

[0232] Incorporation by Reference

[0233] References and citations to other documents, such as patents, patent applications, patent publications, journals, books, papers, web contents, have been made throughout this disclosure. All such documents are hereby incorporated herein by reference in their entirety for all purposes.

[0234] Equivalents

[0235] Various modifications of the invention and many further embodiments thereof, in addition to those shown and described herein, will become apparent to those skilled in the art from the full contents of this document, including references to the scientific and patent literature cited herein. The subject matter herein contains important information, exemplification and guidance that may be adapted to the practice of this invention in its various embodiments and equivalents thereof.

Claims

Claims1. A system for analyzing image data associated with cardiac tissue, the system comprising a hardware processor coupled to non-transitory, computer-readable memory containing instructions executable by the processor to cause the processor to: receive ultrasound image data associated with an anatomical area of interest; and run a tissue analysis algorithm to identify one or more lesions present in tissue at the anatomical area of interest, wherein running the tissue analysis algorithm comprises: processing the data to generate one or more images of tissue at the anatomical area of interest, wherein the one or more images represent cardiac tissue during a complete single heartbeat; analyzing a stress to strain relationship of the tissue; and identifying altered strain in the tissue and detecting one or more lesions in the tissue based, at least in part, on identified altered strain.

2. The system of claim 1, wherein analyzing the stress to strain relationship comprises calculating a stress induced through a displacement of the tissue and a resulting strain of the tissue using one or more of a pattern-based tracking method, a block matching method, and / or tracking of one or more anatomical regions comprising one or more anatomical landmarks and / or structures.

3. The system of claim 2, wherein the images are one or more of two-dimensional (2D) and three-dimensional (3D) volume (2D / 3D volume) images.

4. the system of claim 3, wherein the pattern-based method comprises a speckle tracking analysis within the 2D / 3D volume, wherein a plurality of ultrasound speckles from the 2D / 3D volume are used as input for calculating the stress and the strain.

5. The system of claim 3, wherein the complete single heartbeat is divided into discrete time points such that an individual 2D / 3D volume represents a separate time point within the complete single heartbeat.

6. The system of claim 5, wherein the algorithm calculates the displacement of tissue and the resulting strain based on a movement of the plurality of speckles between time points.

7. The system of claim 3, wherein a displacement between consecutive 2D / 3D volumes are used to calculate the stress and the strain of the tissue.

8. The system of claim 7, wherein a time point when the stress of the tissue is a minimum or a time point when the stress of the tissue is a maximum is calculated from a contraction pattern, wherein the contraction pattern is identified via an electrocardiogram (ECG) or an image based method.

9. The system of claim 3, wherein the algorithm is a machine learning algorithm trained on a plurality of 2D / 3D volume data comprising a plurality of identified tissue displacement in a plurality of complete single heartbeats, wherein a plurality of strongest displacements in the 2D / 3D volume data are identified.

10. The system of claim 9, wherein the machine learning algorithm comprises a neural networkbased feature extractor to predict the tissue displacement from a first 2D / 3D volume to a second 2D / 3D volume.

11. The system of claim 10, wherein the feature extractor uses speckle patterns, and / or one or more other patterns that are distinctive for a region of interest, wherein the speckle patterns and / or the one or more other patterns appear in both the first 2D / 3D volume and the second 2D / 3D volume.

12. The system of claim 11, wherein the algorithm is further trained on labeled training data for end-to-end prediction of strain from two consecutive 2D / 3D volumes.

13. The system of claim 9, wherein the algorithm is further trained using data comprising a plurality of externally measured strain measurements at a known location in the 2D / 3D volume.

14. The system of claim 1, wherein identifying altered strain in the tissue differentiates ablated cardiac tissue from untreated cardiac tissue.

15. The system of claim 14, wherein the algorithm generates a pre-trained model that predicts strain between two ultrasound volumes.

16. The system of claim 15, wherein the pre-trained model further comprises a segmentation task, wherein the segmentation task differentiates ablated cardiac tissue from untreated tissue.

17. The system of claim 15, wherein the model is trained using 2D / 3D volumes of different time points from a plurality of heartbeat cycles to achieve a base for identifying differences in strain between 2D / 3D volumes, wherein the model is trained before ablation.

18. The system of claim 17, wherein a difference in strain between 2D / 3D volumes before ablation and during ablation generates a lesion identification, wherein strain differences are predictive of the lesion extent.

19. The system of claim 18, wherein, the generated lesion identification is output as a volume overlayed on a model of the ultrasound 2D / 3D volume.

20. The system of claim 15, wherein the model is trained during ablation, wherein the model comprises one or more of an unsupervised machine learning model and a supervised model trained with labeled data from one or more of elastography data and histology data identifying lesion tissue.

21. The system of claim 14, wherein the algorithm calculates an ablation depth to generate an ablation depth profile in the tissue over an ablation path.

22. The system of claim 1, wherein the ultrasound data is obtained from a ID or 2D matrix array providing 3D imaging with electronic or mechanic beam steering capabilities.

23. The system of claim 22, wherein the received ultrasound image data comprises full circumferential 3D image data obtained from an intracardiac echography (ICE) catheter, wherein the ultrasound imaging data is real-time ultrasound image data comprising 3D volume data associated with cardiac tissue.

24. The system of claim 23, wherein the data comprises one or more of a plurality of time- sequentially acquired B-Images, beamformed RF data, and enveloped data over a full perimeter of a catheter tip, wherein the data corresponds to a cross-section of the tissue during the complete single heartbeat.

25. The system of claim 23, wherein the ultrasound image data comprises a complete simultaneously acquired volume around the full perimeter of a catheter tip.

26. The system of claim 1, wherein the system is further configured to preprocess the ultrasound image data, wherein preprocess comprises one or more of noise removal, image smoothing, and contrast enhancement.

27. The system of claim 1, wherein analyzing further comprises analyzing data from one or more ultrasound signals at one or more signal stages, wherein the one or more signal stages comprise compounded, envelope, and log-compressed.

28. The system of claim 1, wherein analyzing further comprises analyzing data from one or more methods comprising a Nakagami method, and a microstructural envelop statistics method.

29. The system of claim 1, wherein analyzing further comprises analyzing data related to one or more of perfusion, stiffness, anisotropy, coherence, specific statistical distributions in tissue, frequency power spectrum of tissue, ultrasound data at various beamforming stages, one or more learned features from raw and / or processed signals, interventional tool tracking data, and breath data.

30. A method for analyzing cardiac tissue images, the method comprising:receiving ultrasound image data associated with an anatomical area of interest; and running a tissue analysis algorithm to identify one or more lesions present in tissue at the anatomical area of interest, wherein running the tissue analysis algorithm comprises: processing the data to generate one or more images of tissue at the anatomical area of interest, wherein the one or more images represent cardiac tissue during a complete single heartbeat; analyzing a stress to strain relationship of the tissue; and identifying altered strain in the tissue and detecting one or more lesions in the tissue based, at least in part, on identified altered strain.

31. The method of claim 30, wherein analyzing the stress to strain relationship comprises calculating a stress induced through a displacement of the tissue and a resulting strain of the tissue using one or more of a pattern-based tracking method, a block matching method, and / or tracking of one or more anatomical regions comprising one or more anatomical landmarks and / or structures.

32. The method of claim 31, wherein the images are one or more of two-dimensional (2D) and three-dimensional (3D) volume (2D / 3D volume) images.

33. The method of claim 32, wherein the pattern-based method comprises a speckle tracking analysis within the 3D volume, wherein a plurality of ultrasound speckles from the 2D / 3D volume are used as input for calculating the stress and the strain.

34. The method of claim 33, wherein the complete single heartbeat is divided into discrete time points such that an individual 2D / 3D volume represents a separate time point within the complete single heartbeat.

35. The method of claim 34, wherein the algorithm calculates the displacement of tissue and the resulting strain based on a movement of the plurality of speckles between time points.

36. The method of claim 32, wherein a displacement between consecutive 2D / 3D volumes are used to calculate the stress and the strain of the tissue.

37. The method of claim 36, wherein a time point when the stress of the tissue is a minimum or a time point when the stress of the tissue is a maximum is calculated from a contraction pattern, wherein the contraction pattern is identified via an electrocardiogram (ECG) or an image-based method.

38. The method of claim 32, wherein the algorithm is a machine learning algorithm trained on a plurality of 2D / 3D volume data comprising a plurality of identified tissue displacement in a plurality of complete single heartbeats, wherein a plurality of strongest displacements in the 2D / 3D volume data are identified.

39. The method of claim 38, wherein the machine learning algorithm comprises a neural network-based feature extractor to predict the tissue displacement from a first 2D / 3D volume to a second 2D / 3D volume.

40. The method of claim 39, wherein the feature extractor uses speckle patterns, and / or one or more other patterns that are distinctive for a region of interest, wherein the speckle patterns and / or the one or more other patterns appear in both the first 2D / 3D volume and the second 2D / 3D volume.

41. The method of claim 40, wherein the algorithm is further trained on labeled training data for end-to-end prediction of strain from two consecutive 2D / 3D volumes.

42. The method of claim 38, wherein the algorithm is further trained using data comprising a plurality of externally measured strain measurements at a known location in the 2D / 3D volume.

43. The method of claim 32, wherein identifying altered strain in the tissue differentiates ablated cardiac tissue from untreated cardiac tissue.

44. The method of claim 43, wherein the algorithm generates a pre-trained model that predicts strain between two ultrasound volumes.

45. The method of claim 44, wherein the pre-trained model further comprises a segmentation task, wherein the segmentation task differentiates ablated cardiac tissue from untreated tissue.

46. The method of claim 45, wherein the model is trained using 2D / 3D volumes of different time points from a plurality of heartbeat cycles to achieve a base for identifying differences in strain between 2D / 3D volumes, wherein the model is trained before ablation.

47. The method of claim 45, wherein the model is trained during ablation, wherein the model comprises one or more of an unsupervised machine learning model and a supervised model trained with labeled data from one or more of elastography data and histology data identifying lesion tissue.

48. The method of claim 44, wherein a difference in strain between 2D / 3D volumes before ablation and during ablation generates a lesion identification, wherein strain differences are predictive of the lesion extent.

49. The method of claim 48, wherein, the generated lesion identification is output as a volume overlayed on a model of the 2D / 3D volume.

50. The method of claim 43, wherein the algorithm calculates an ablation depth to generate an ablation depth profile in the tissue over an ablation path.

51. The method of claim 30, wherein the ultrasound data is obtained from a ID or 2D matrix array providing 3D imaging with electronic or mechanic beam steering capabilities.

52. The method of claim 51, wherein the received ultrasound image data comprises full circumferential 3D image data obtained from an intracardiac echography (ICE) catheter, whereinthe ultrasound imaging data is real-time ultrasound image data comprising 3D volume data associated with cardiac tissue.

53. The method of claim 52, wherein the data comprises one or more of a plurality of time- sequentially acquired B-Images, beamformed RF data, and enveloped data over a full perimeter of a catheter tip, wherein the data corresponds to a cross-section of the tissue during the complete single heartbeat.

54. The method of claim 52, wherein the ultrasound image data comprises a complete simultaneously acquired volume around the full perimeter of a catheter tip.

55. The method of claim 30, wherein the system is further configured to preprocess the ultrasound image data, wherein preprocess comprises one or more of noise removal, image smoothing, and contrast enhancement.

56. The method of claim 30, wherein analyzing further comprises analyzing data from one or more ultrasound signals at one or more signal stages, wherein the one or more signal stages comprise compounded, envelope, and log-compressed.

57. The method of claim 30, wherein analyzing further comprises analyzing data from one or more methods comprising a Nakagami method, and a microstructural envelop statistics method.

58. The method of claim 30, wherein analyzing further comprises analyzing data related to one or more of perfusion, stiffness, anisotropy, coherence, specific statistical distributions in tissue, frequency power spectrum of tissue, ultrasound data at various beamforming stages, one or more learned features from raw and / or processed signals, interventional tool tracking data, and breath data.

Images