Method and system for predicting abdominal aortic aneurysm (AAA) growth

Abstract

A method, system, and non-transitory storage medium are provided for predicting abdominal aortic aneurysm (AAA) growth in a patient diagnosed with an AAA. A segmented region of interest (ROI) including the aorta and adjacent structures is received by segmenting a set of images. Wall shear stress and luminal thickness parameters are determined. A 3D parametric mesh is generated including multiple concentric 3D mesh layers, each including the same predetermined number of nodes. The generation includes encoding the segmented ROI, wall shear stress, and luminal thickness parameters as features at individual node locations of the 3D parametric mesh. A trained growth prediction machine learning model predicts whether a particular patient will exhibit AAA growth based on at least a subset of the features of the 3D parametric mesh. Training of the growth prediction model is also disclosed.

Description

[Technical Field]

[0001] The present technology relates to the field of medical imaging, and more precisely, to methods, systems, and non-transitory computer-readable media for predicting abdominal aortic aneurysm (AAA) growth. [Background technology]

[0002] Post-processing of vascular imaging helps to quantify multiple clinically relevant variables, enhancing the diagnostic and prognostic utility of the scan to the interpreting radiologist (reader) or referring physician. Specifically, this collection of variables generated from the images constitutes a more comprehensive assessment of vascular function, resulting in a more holistic assessment of the patient's health status and more patient-specific treatment approaches and risk assessments.

[0003] A variety of information can be obtained from routinely obtaining three-dimensional imaging of the abdominal aorta, consisting of either multiphase or static images. Such imaging is often prescribed at the time of diagnosis, to confirm the diagnosis of abdominal aortic aneurysm (AAA), or may aid in the diagnosis of this disease. At the time of diagnosis, the problem of predicting the progression of AAA remains unsolved. Using the acquired images, the region involved in the AAA can be isolated (segmented) and a three-dimensional representation of the aneurysm can be obtained (meshing). The mesh can be used to calculate various geometric and functional markers, such as fluid dynamic estimators, kinematic and biomechanical quantification, or structural features.

[0004] The quantified variables can come from multiple physics domains, such as structural mechanics (e.g., geometry, shape, and deformation), fluid mechanics (e.g., luminal flow), or descriptive variables obtained directly from image processing (e.g., image intensity-related variables). Each of these domains must be processed according to a specific data format, such as a shell mesh, a 3D solid mesh, or array-like data. These data formats can come at different data densities from multiple domains, different patients, and often from subsequent scans performed on the same patient. These differences in data formats and local data densities hinder the ability to systematically report all available information from the various domains as independent or combined variables for each region of the aorta.

[0005] Furthermore, when training diagnostic or prognostic machine learning models, heterogeneity in data formats forces data scientists to define multiple models with more parameters, and this latter aspect requires more subjects to adequately train generalizable diagnostic and predictive models. Summary of the Invention

[0006] The objective of the present technology is to ameliorate at least some of the inconveniences present in the prior art. One or more embodiments of the present technology may provide and / or expand a range of approaches and / or methods for achieving the goals and objectives of the technology.

[0007] One or more embodiments of the present technology are developed based on the developer's recognition that prognostic and diagnostic applications require a large number of variables or biomarkers to be obtained from multiple physical domains and descriptive representations, which may require data obtained from multiple imaging modalities, a combination of models, simulations, and other computational techniques. Each of these variables can be obtained via a different data structure with different data densities and different data formats for the region of the anatomy under investigation. However, these various representations may not be ideal for clinical reporting and training machine learning models.

[0008] The developers of this technology propose a solution that allows homogenizing data formats and domain data densities across data domains, between and across patients, facilitating multi-domain reporting and modeling for machine learning applications.

[0009] One or more embodiments of the present technology provide an anatomically relevant meshing strategy to produce homogenized data across multiple modalities and scans. Parametric meshes generated using the present disclosure allow data from all data types, ranging from shell and solid meshes to array-like data including pixel-specific data, to be stored in stackable layers that are easy to interpret and utilize, relying on a single type of data encoding, for training more compact machine learning-based models.

[0010] One or more embodiments of the present technology provide a 3D parametric mesh comprising multiple concentric three-dimensional (3D) mesh layers, where each concentric 3D mesh layer can be interpreted as a separate 3D mesh representing a different interior or exterior layer of an anatomical structure of interest, thereby allowing multi-domain information regarding the inner layer (e.g., centerline geometry), outer layer (e.g., wall geometry, distortion), and middle layer (e.g., lumen geometry, blood flow) of the anatomical structure to be stored and visually represented in the 3D parametric mesh.

[0011] One or more embodiments of the present method and system convert multiple vessel-specific data types into multi-channel, anatomically relevant, stackable images that can be used to train diagnostic and prognostic artificial intelligence-based models, as well as provide organized, intuitive multi-domain reports to medical professionals.

[0012] One or more embodiments of the present technology enable modular modeling for diagnostic and prognostic purposes, leveraging each available domain of information. Because all models rely on the same data types, weights can be optionally shared or very minimally retrained when new information domains are introduced. Modular modeling makes it possible to retrain new architectures or for new tasks by utilizing fewer weights (parameters) and requiring less retraining of these weights. Consequently, this makes it easier to obtain generalizable models starting with a smaller number of vascular scans.

[0013] Accordingly, one or more embodiments of the present technology are directed to methods and systems for training and using machine learning models to predict the growth of aneurysms in blood vessels, such as AAAs.

[0014] According to a broad aspect of the present technology, there is provided a method for generating a 3D parametric mesh of an anatomical structure and storing multi-domain data therein, the method being executed by at least one processor. The method includes: receiving a plurality of anatomical segments of at least some of the anatomical structures within a given patient body obtained from segmentation of an image set acquired by a medical imaging device, the image set including at least one image of at least some of the anatomical structures within the given patient body; receiving a 3D mesh for representing the anatomical structures, the 3D mesh including a plurality of concentric 3D mesh layers, each of the plurality of concentric 3D mesh layers including the same predetermined number of nodes; determining at least one distinct set of nodes in the 3D mesh corresponding to at least one distinct anatomical segment of the plurality of anatomical segments and obtaining a distinct correspondence rule therebetween; and encoding the at least one set of nodes in the 3D mesh with a distinct set of features from the at least one distinct anatomical segment using the correspondence rule to obtain a 3D parametric mesh, wherein each node of the at least one set of nodes in the 3D parametric mesh is associated with a distinct plurality of feature channels including a distinct feature set.

[0015] In one or more embodiments of the method, at least a subset of nodes of the at least one set of nodes are located on different concentric 3D mesh layers.

[0016] In one or more embodiments of the method, encoding at least one node set of the 3D mesh with a distinct feature set from at least one distinct anatomical segment using a correspondence rule to obtain a 3D parametric mesh includes determining a distinct feature set from biomarkers in the at least one distinct anatomical segment using a distinct correspondence rule, and assigning each of the at least one distinct node set with a distinct feature set from the at least one distinct anatomical segment.

[0017] In one or more embodiments of the method, the method further includes receiving a domain representation including individual biomarkers associated with anatomical structures within a given patient body; determining at least one other individual set of nodes in the 3D parametric mesh corresponding to at least one other region in the domain representation to obtain another individual correspondence rule, wherein at least a subset of the other individual second set of nodes is located on a different concentric 3D mesh layer; and encoding the at least one other individual set of nodes in the 3D parametric mesh with another feature set based on the individual biomarkers using the other individual correspondence rule, wherein each node of the at least one other individual set of nodes in the 3D parametric mesh is associated with a plurality of individual feature channels including the other feature set.

[0018] In one or more implementations of the method, each individual node is further associated with at least one time frame for representing the 3D parametric mesh in time.

[0019] In one or more implementations of the method, each distinct concentric 3D mesh layer is represented as a distinct multidimensional array, and the position of a given node on the distinct 3D mesh layer corresponds to the position of a given node in the distinct multidimensional array.

[0020] In one or more implementations of the method, the multiple feature channels for each node of a separate 3D mesh layer are represented as a separate node array, with each cell of the separate node array corresponding to a separate feature channel of the multiple feature channels.

[0021] In one or more implementations of the method, the domain representation includes another mesh that is different from the 3D parametric mesh, and the other individual correspondence rule includes determining a mapping between nodes in the other mesh and nodes in the 3D parametric mesh.

[0022] In one or more embodiments of the method, the other mesh comprises one of a polygonal mesh, the polygonal mesh comprising one of a triangular mesh, a quadrilateral mesh, a convex polygonal mesh, a concave polygonal mesh, and a polygonal mesh with holes.

[0023] In one or more embodiments of the method, the domain representation includes a structural mechanics representation, and the individual biomarkers include structural mechanics biomarkers, which include at least one of pressure values, strain values, and deformation values.

[0024] In one or more embodiments of the method, the domain representation includes a fluid dynamics representation and the individual biomarkers include at least one of a blood flow value and a wall shear stress value.

[0025] In one or more embodiments of the method, the domain representation comprises a descriptive variable representation and the individual biomarkers comprise at least one of geometric data values ​​and image data values.

[0026] In one or more embodiments of the method, the anatomical structure includes the aorta of a given patient.

[0027] In one or more embodiments of the method, the plurality of anatomical segments includes the lumen and the aortic wall.

[0028] In accordance with a broad aspect of the present technology, there is provided a system for generating a 3D parametric mesh of an anatomical structure and storing multi-domain data therein, the system comprising at least one processor and a non-transitory storage medium operatively connected to the at least one processor, the non-transitory storage medium storing computer-readable instructions. The at least one processor, when executing the computer-readable instructions, is configured to: receive a plurality of anatomical segments of at least some structures of an anatomical structure within a given patient body obtained from segmentation of an image set acquired by a medical imaging device, the image set including at least one image of at least some structures of an anatomical structure within the given patient body; receive a 3D mesh for representing the anatomical structures, the 3D mesh including a plurality of concentric 3D mesh layers, each of the plurality of concentric 3D mesh layers including the same predetermined number of nodes; determine at least one distinct set of nodes in the 3D mesh corresponding to at least one distinct anatomical segment of the plurality of anatomical segments to obtain a distinct correspondence rule therebetween; and encode the at least one set of nodes of the 3D mesh with a distinct set of features from the at least one distinct anatomical segment using the correspondence rule to obtain a 3D parametric mesh, wherein each node of the at least one set of nodes in the 3D parametric mesh is associated with a distinct plurality of feature channels including a distinct feature set.

[0029] In one or more embodiments of the system, at least a subset of nodes of the at least one set of nodes are located on different concentric 3D mesh layers.

[0030] In one or more embodiments of the system, encoding at least one set of nodes of the 3D mesh with a distinct set of features from at least one distinct anatomical segment using a correspondence rule to obtain a 3D parametric mesh includes determining a distinct set of features from biomarkers in the at least one distinct anatomical segment using a distinct correspondence rule, and assigning each of the at least one distinct set of nodes with a distinct set of features from the at least one distinct anatomical segment.

[0031] In one or more embodiments of the system, the at least one processor is further configured to: receive a domain representation including individual biomarkers associated with anatomical structures within a given patient body; determine at least one other individual set of nodes in the 3D parametric mesh corresponding to at least one other region in the domain representation to obtain another individual correspondence rule, wherein at least a subset of the other individual second set of nodes is located on a different concentric 3D mesh layer; and encode the at least one other individual set of nodes in the 3D parametric mesh with another feature set based on the individual biomarkers using the other individual correspondence rule, wherein each node of the at least one other individual set of nodes in the 3D parametric mesh is associated with a plurality of individual feature channels including the other feature set.

[0032] In one or more embodiments of the system, each individual node is further associated with at least one time frame for representing the 3D parametric mesh in time.

[0033] In one or more embodiments of the system, each individual concentric 3D mesh layer is represented as an individual multidimensional array, and the position of a given node on the individual 3D mesh layer corresponds to the position of a given node in the individual multidimensional array.

[0034] In one or more embodiments of the system, the multiple feature channels for each node of a separate 3D mesh layer are represented as a separate node array, with each cell of the separate node array corresponding to a separate feature channel of the multiple feature channels.

[0035] In one or more embodiments of the system, the domain representation includes another mesh that is different from the 3D parametric mesh, and another separate correspondence rule includes determining a mapping between nodes in the another mesh and nodes in the 3D parametric mesh.

[0036] In one or more embodiments of the system, the other mesh includes one of a polygonal mesh, the polygonal mesh including one of a triangular mesh, a quadrilateral mesh, a convex polygonal mesh, a concave polygonal mesh, and a polygonal mesh with holes.

[0037] In one or more embodiments of the system, the domain representation includes a structural mechanics representation, and the individual biomarkers include structural mechanics biomarkers, which include at least one of pressure values, strain values, and deformation values.

[0038] In one or more embodiments of the system, the domain representation includes a fluid dynamics representation and the individual biomarkers include at least one of a blood flow value and a shear stress value.

[0039] In one or more embodiments of the system, the domain representation includes a descriptive variable representation and the individual biomarkers include at least one of geometric data values ​​and image data values.

[0040] In one or more embodiments of the system, the anatomical structure includes the aorta of a given patient.

[0041] In one or more embodiments of the system, the plurality of anatomical segments includes the lumen and the aortic wall.

[0042] According to a broad aspect of the present technology, there is provided a method for predicting abdominal aortic aneurysm (AAA) growth based on at least one image of a given patient previously diagnosed with AAA, the method being executed by at least one processor, the at least one processor having access to a trained growth prediction machine learning (ML) model, the method including: receiving a baseline image set of a body including an aorta of the given patient, the image set including at least one image, the image set being acquired using a medical imaging device; segmenting the image set using the at least one trained segmentation model to obtain a segmented region of interest (ROI) of the aorta and adjacent structures; and performing a predictive analysis based on the segmented ROI of the aorta. generating a wall shear stress parameter and determining an intraluminal thickness parameter based on a segmented ROI of the aorta; generating a 3D parametric mesh based on the segmented ROI of the aorta, the 3D parametric mesh including multiple concentric 3D mesh layers, each of the multiple concentric 3D mesh layers including the same predetermined number of nodes, wherein the generating includes encoding the segmented ROI, the wall shear stress parameter, and the intraluminal thickness parameter as features at individual node locations; and predicting whether a given patient will exhibit AAA growth using the trained growth prediction ML model based on at least a subset of the features of the 3D parametric mesh.

[0043] In one or more embodiments of the method, the ROI of the aorta and adjacent structures includes the abdominal aortic region and the iliac arteries.

[0044] In one or more embodiments of the method, the ROI of the aorta and adjacent structures further includes a portion of the spine.

[0045] In one or more embodiments of the method, generating a parametric mesh based on the segmented ROI of the aorta includes encoding pixel locations and pixel intensity values ​​at individual node locations.

[0046] In one or more embodiments of the method, the subset of features includes geometric features, where the geometric features include 2D distances to a centerline of the parametric mesh.

[0047] In one or more embodiments of the method, the geometric feature comprises a 3D distance to a centerline of the parametric mesh.

[0048] In one or more embodiments of the method, the method further includes, prior to receiving the baseline image set, training the growth prediction model with a training dataset to obtain a trained growth prediction model, the training dataset including, for each individual patient of the plurality of patients, an individual comparison of encoded features between a baseline mask and a follow-up mask of an individual 3D parametric mesh generated for the individual patient based on an individual baseline image set and an individual follow-up image set, and an individual growth label indicating the presence of AAA growth.

[0049] In one or more embodiments of the method, the individual comparison of the encoded features includes an individual comparison of diameters at individual sections of nodes perpendicular to individual aortic centerlines of the individual parametric meshes.

[0050] In one or more embodiments of the method, the individual comparison of the encoded features includes a comparison of diameters at individual sections of the nodes perpendicular to the individual aortic centerlines of the individual parametric meshes.

[0051] According to a broad aspect of the present technology, there is provided a method for predicting growth of an aneurysm in a blood vessel based on at least one image of a given patient previously diagnosed with an aneurysm, the method being executed by at least one processor, the at least one processor having access to a trained growth prediction machine learning (ML) model, the method comprising: receiving segmented regions of interest (ROIs) of a blood vessel and adjacent structures segmented from a set of images of the given patient acquired using a medical imaging device; generating wall shear stress parameters based on the segmented ROIs of the blood vessel; and generating endoluminal thickness parameters based on the segmented ROIs of the blood vessel. generating a 3D parametric mesh based on the segmented ROI of the blood vessel, the 3D parametric mesh including multiple concentric 3D mesh layers, each of the multiple concentric 3D mesh layers including the same predetermined number of nodes, wherein generating includes encoding the segmented ROI, wall shear stress parameters, and luminal thickness parameters as features at individual node locations; and predicting whether a given patient will exhibit aneurysm growth using the trained growth prediction ML model based on at least a subset of the features of the 3D parametric mesh.

[0052] According to a broad aspect of the present technology, there is provided a system for predicting abdominal aortic aneurysm (AAA) growth based on at least one image of a given patient previously diagnosed with AAA, the system comprising: a non-transitory storage medium storing computer-readable instructions; and at least one processor operatively connected to the non-transitory storage medium, the at least one processor having access to a trained growth prediction ML model. The at least one processor, when executing the computer-readable instructions, is configured to: receive a baseline image set of a body including the aorta of a given patient, the image set including at least one image, the image set acquired using a medical imaging device; segment the image set using at least one trained segmentation model to obtain segmented regions of interest (ROIs) of the aorta and adjacent structures; generate a wall shear stress parameter based on the segmented ROI of the aorta; determine an endoluminal thickness parameter based on the segmented ROI of the aorta; generate a 3D parametric mesh based on the segmented ROI of the aorta, the 3D parametric mesh including multiple concentric 3D mesh layers, each of the multiple concentric 3D mesh layers including the same predetermined number of nodes, the generating including encoding the segmented ROI, the wall shear stress parameter, and the endoluminal thickness parameter as features at individual node locations; and predict whether the given patient will exhibit AAA growth using the trained growth prediction model based on at least a subset of the features of the parametric mesh.

[0053] In one or more embodiments of the system, the ROI of the aorta and adjacent structures includes the abdominal aortic region and the iliac arteries.

[0054] In one or more embodiments of the system, the ROI of the aorta and adjacent structures includes a portion of the spine.

[0055] In one or more embodiments of the method, generating a parametric mesh based on the segmented ROI of the aorta includes encoding pixel locations and pixel intensity values ​​at individual node locations.

[0056] In one or more embodiments of the system, the subset of features includes geometric features, where the geometric features include 2D distances to a centerline of the parametric mesh.

[0057] In one or more embodiments of the system, the geometric feature includes a 3D distance to a centerline of the parametric mesh.

[0058] In one or more embodiments of the system, the at least one processor is further configured to, prior to receiving the baseline image set, train the growth prediction model with a training dataset to obtain a trained growth prediction model, the training dataset including, for each individual patient of the plurality of patients, an individual comparison of encoded features between a baseline mask and a follow-up mask of an individual 3D parametric mesh generated for the individual patient based on the individual baseline image set and the individual follow-up image set, and an individual growth label indicating the presence of AAA growth.

[0059] In one or more embodiments of the system, the individual comparison of the encoded features includes a comparison of diameters at individual sections of nodes perpendicular to the individual aortic centerlines of the individual parametric meshes.

[0060] In one or more embodiments of the system, the individual comparison of the encoded features includes a comparison of diameters at individual sections of nodes perpendicular to the individual aortic centerlines of the individual parametric meshes.

[0061] In accordance with a broad aspect of the present technology, there is provided a system for predicting aneurysm growth based on at least one image of a given patient previously diagnosed with an aneurysm, the system comprising: a non-transitory storage medium storing computer-readable instructions; and at least one processor operatively connected to the non-transitory storage medium, the at least one processor having access to a trained growth-prediction machine learning (ML) model, the at least one processor, when executed, to: receive segmented regions of interest (ROIs) of blood vessels and adjacent structures from a set of images of the given patient acquired using a medical imaging device; and generate wall shear stress parameters based on the segmented ROIs of the blood vessels. determining an intraluminal thickness parameter based on the segmented ROI of the blood vessel; generating a 3D parametric mesh based on the segmented ROI of the blood vessel, the 3D parametric mesh including multiple concentric 3D mesh layers, each of the multiple concentric 3D mesh layers including the same predetermined number of nodes, the generating including encoding the segmented ROI, the wall shear stress parameter, and the intraluminal thickness parameter as features at individual node locations; and predicting whether a given patient will exhibit aneurysm growth using the trained growth prediction ML model based on at least a subset of the features of the 3D parametric mesh.

[0062] Terms and Definitions In the context of this specification, a "server" is a computer program that runs on appropriate hardware and can receive requests (e.g., from electronic devices) over a network (e.g., a communications network) and execute those requests or cause those requests to be executed. The hardware can be a physical computer or a physical computer system, but need not be either, in the context of the present technology. In this context, the use of the term "server" is not intended to imply that all tasks (e.g., received instructions or requests) or any particular task will be received, executed, or caused to be executed by the same server (i.e., the same software and / or hardware). It is intended to mean that any number of software elements or hardware devices may be involved in receiving / sending, executing, or causing to be executed any task or request, or the result of any task or request. All of this software and hardware may be one server or multiple servers, both of which are included in the terms "at least one server" and "server."

[0063] In the context of this specification, an "electronic device" is any computing device or computer hardware capable of executing software appropriate for the relevant task at hand. Thus, some (non-limiting) examples of electronic devices include general-purpose personal computers (desktops, laptops, netbooks, etc.), mobile computing devices, smartphones, and tablets, as well as network equipment such as routers, switches, and gateways. Note that an electronic device in this context does not exclude acting as a server to other electronic devices. The use of the expression "electronic device" does not exclude multiple electronic devices from being used to receive / send, execute, or cause to be executed a task or request, or the results of a task or request, or any method step, described herein. In the context of this specification, a "client device" refers to any of a variety of end-user client electronic devices associated with a user, such as a personal computer, tablet, smartphone, etc.

[0064] In the context of this specification, unless expressly specified otherwise, computer system refers to, but is not limited to, an "electronic device," "client device," "computing device," "operating system," "system," "computer-based system," "computer system," "network system," "network device," "controller unit," "monitoring device," "control device," "server," and / or any combination thereof appropriate to the task at hand.

[0065] In the context of this specification, the expression "computer-readable storage medium" (also referred to as "storage medium" and "storage") is intended to include non-transitory media of whatever nature and type, including but not limited to RAM, ROM, disks (CD-ROM, DVD, floppy disk, hard drive, etc.), USB keys, solid state drives, tape drives, etc. A computer information storage medium may be formed by combining multiple components, including two or more media components of the same type and / or two or more media components of different types.

[0066] In the context of this specification, a "database" is any structured collection of data, regardless of its particular structure, database management software, or computer hardware on which the data is stored, implemented, or otherwise made available. A database may reside on the same hardware as the processes that store or utilize the information stored in the database, or it may reside on separate hardware, such as a dedicated server or multiple servers.

[0067] In the context of this specification, the expression "information" includes information of any nature or type that can be stored in a database, and thus includes, but is not limited to, audiovisual works (images, films, sound recordings, presentations, etc.), data (location data, numerical data, etc.), text (opinions, comments, questions, messages, etc.), documents, spreadsheets, word lists, etc.

[0068] In the context of this specification, unless expressly specified otherwise, a "representation" of an information element may be the information element itself, or it may be a pointer, reference, link, or other indirection mechanism that allows the recipient of the representation to locate a network, memory, database, or other computer-readable medium where the information element can be obtained. For example, a representation of a document may include the document itself (i.e., its contents), a unique document descriptor that identifies the file with respect to a particular file system, or other means that directs the recipient of the representation to a network location, memory address, database table, or other location where the file can be accessed. As those skilled in the art will recognize, the degree of precision required in such a representation depends on the degree of prior understanding of any interpretation to be given to the information exchanged between the sender and recipient of the representation. For example, if, prior to communication between the sender and recipient, a representation of an information element is known to take the form of a database key for an entry in a particular table of a given database that contains the information element, then merely transmitting the database key can effectively convey the information element to the recipient, even though the information element itself has not been transmitted between the sender and recipient of the representation.

[0069] In the context of this specification, the expression "communications network" is intended to include telecommunications networks such as computer networks, the Internet, telephone networks, Telex networks, TCP / IP data networks (e.g., WAN networks, LAN networks, etc.), etc. The term "communications network" includes wired networks or direct-wired connections, wireless media such as acoustic, radio frequency (RF), infrared, and other wireless media, and combinations of any of the above.

[0070] In the context of this specification, words such as “first,” “second,” and “third” are used as adjectives solely to distinguish between the nouns they modify, and not to describe a particular relationship between those nouns. Thus, for example, it should be understood that the use of the terms “first server” and “third server” is not intended to suggest a particular order, type, chronology, hierarchy, or (e.g.) ranking between servers, nor is their use (in and of itself) intended to suggest that a “second server” must necessarily be present in any given situation. Furthermore, as described elsewhere herein, reference to a “first” element and a “second” element does not exclude the two elements from being the same actual, real-world element. Thus, for example, in some cases, the “first” server and the “second” server may be the same software and / or hardware, while in other cases, they may be different software and / or hardware.

[0071] Each embodiment of the present technology will have at least one, but not necessarily all, of the above-described objects and / or aspects, and it will be understood that some aspects of the present technology that arise from seeking to achieve the above-described object may not meet that object and / or may meet other objects not specifically set forth herein.

[0072] Additional and / or alternative features, aspects, and advantages of embodiments of the present technology will become apparent from the following description, the accompanying drawings, and the appended claims.

[0073] For a better understanding of the present technology, as well as other aspects and further features thereof, reference is made to the following description taken in conjunction with the accompanying drawings. [Brief explanation of the drawings]

[0074] [Figure 1]1 shows a schematic diagram of an electronic device in accordance with one or more non-limiting embodiments of the present technology. [Figure 2] 1 shows a schematic diagram of a communication system in accordance with one or more non-limiting embodiments of the present technology. [Figure 3] FIG. 1 shows a schematic diagram of a parametric mesh generation procedure in accordance with one or more non-limiting embodiments of the present technique. [Figure 4] FIG. 1 shows a non-limiting example of a perspective view of a visual rendering of a first parametric mesh taken from the anterior left side of the aorta including the iliac arteries, in accordance with one or more non-limiting embodiments of the present technology. [Figure 5A] 10 shows a perspective view of a visual rendering of the parametric mesh of FIG. 4 with the top portion removed according to line 11 to reveal multiple concentric 3D mesh layers. [Figure 5B] 5B shows a detailed view of a visual rendering of the parametric mesh of FIG. 5A, along with a schematic view of a selected node and its multiple feature channels. [Figure 6] A top view of a visual rendering of the parametric mesh 400 of FIG. 5 with the upper portion removed according to line 11 is shown, along with a schematic diagram of a selected node and its multiple feature channels. [Figure 7] FIG. 10 shows a perspective view of a visual rendering of a second parametric mesh taken from the anterior left side of the aorta and iliac arteries, in accordance with one or more non-limiting embodiments of the present technology. [Figure 8] FIG. 10 shows a schematic diagram of a first user interface illustrating a visual representation of a third parametric mesh of the aorta including the iliac arteries and a corresponding outer layer arrangement with user interface components for navigating information within the third parametric mesh. [Figure 9A]FIG. 10 shows a schematic diagram of a second user interface showing a visual representation of a fourth parametric mesh of an aorta with iliac arteries with concentric mesh layers between the lumen and the wall selected and a corresponding layer arrangement with user interface components for navigating information within the fourth parametric mesh, the second user interface being shown in accordance with one or more non-limiting embodiments of the present technology. [Figure 9B] 9B shows a schematic diagram of the second user interface of FIG. 9A with the core concentric mesh layer selected in the user interface component. [Figure 10A] 1 shows a flowchart of a method for generating a parametric mesh performed in accordance with one or more non-limiting embodiments of the present technique. [Figure 10B] 1 shows a flowchart of a method for generating a parametric mesh performed in accordance with one or more non-limiting embodiments of the present technique. [Figure 11] 1 shows a schematic diagram of a AAA growth prediction training procedure according to a non-limiting embodiment of the present technology. [Figure 12] 10A-10C illustrate non-limiting examples of baseline and follow-up mask views and comparisons of baseline and follow-up masks of a parametric mesh, in accordance with non-limiting implementations of the present technology. [Figure 13] 10 illustrates different inputs and outputs of a parametric mesh generation procedure used to perform an AAA growth prediction training procedure, in accordance with a non-limiting embodiment of the present technology. [Figure 14] Non-limiting examples of distributions of TAWSS (A), wall strain (B), and ILT (C) on the lumen surface of a parametric mesh are shown. Each panel shows the nodal distribution of the variables on the left and patch-based encoding for local characterization on the right. [Figure 15]Figure 1 shows a non-limiting example of regional growth assessed as a measure of regional diameter change, determined by aligning the reconstructed geometry at baseline and follow-up on a parametric mesh and comparing the diameters at multiple sections perpendicular to the aortic centerline, shown as a black line along the length of the aorta. [Figure 16] Figure 1 shows SHAP dependency plots illustrating the effect of each biomechanics-based biomarker on growth prediction: TAWSS (A), strain (B), and ILT (C), as well as maximum aortic diameter at baseline (D) in one non-limiting example of an Extra Trees classification model trained to perform growth prediction. [Figure 17] Figure 1 shows a plot of the receiver operating characteristic (ROC) curve for the ExtraTree classification model with the area under the curve (AUC) reported. The ExtraTree algorithm was used as a binary classifier, with the positive class representing patches with diameter growth ≥ 2.5 mm / year. [Figure 18] 1 shows a SHAP summary plot illustrating the importance of all features contributing to model prediction in one non-limiting example of a model trained to perform growth prediction. [Figure 19] 1 shows a flowchart of a method for generating training data using a parametric mesh, the method being performed in accordance with one or more non-limiting implementations of the present technology. [Figure 20] 1 shows a flowchart of a method for training a model to perform growth prediction based on training data generated from a parametric mesh, the method being performed in accordance with one or more non-limiting implementations of the present technology. [Figure 21] 1 shows a flowchart of a method for performing growth prediction using a trained model, the method being performed in accordance with one or more non-limiting implementations of the present technology. DETAILED DESCRIPTION OF THE INVENTION

[0075] The examples and conditional language described herein are intended primarily to aid the reader in understanding the principles of the present technology, and are not intended to limit the scope of the present technology to such specifically described examples and conditions. Those skilled in the art will understand that, although not explicitly described or shown herein, they can devise various configurations that embody the principles of the present technology and are within its spirit and scope.

[0076] Furthermore, to aid in understanding, the following description may describe a relatively simplified implementation of the technology, as those skilled in the art will appreciate that various implementations of the technology can be more complex.

[0077] In some cases, what are believed to be useful examples of modifications to the technology may also be described. This is done merely to aid in understanding and, again, does not define the scope or delimit the technology. These modifications are not an exhaustive list, and one of ordinary skill in the art may nonetheless make other modifications while remaining within the scope of the technology. Furthermore, if examples of modifications are not described, it should not be construed that modifications are not possible and / or that what is described is the only way to implement that element of the technology.

[0078] Furthermore, all statements herein reciting principles, aspects, and implementations of the present technology, as well as specific examples thereof, are intended to encompass both structural and functional equivalents, whether now known or developed in the future. Thus, for example, those skilled in the art will understand that any block diagrams herein represent conceptual views of illustrative circuitry embodying the principles of the present technology. Similarly, flowcharts, flow diagrams, state transition diagrams, pseudocode, and the like, may be substantially embodied on computer-readable media and will be understood to represent various processes that may be executed by such a computer or processor, whether or not such a computer or processor is explicitly shown.

[0079] The functionality of the various elements shown in the figures may be provided through the use of dedicated hardware and hardware capable of executing software in conjunction with appropriate software, including any functional blocks labeled as a "processor" or "graphics processing unit." When provided by a processor, the functionality may be provided by a single dedicated processor, by a single shared processor, or by multiple individual processors, some of which may be shared. In some non-limiting implementations of the present technology, the processor may be a general-purpose processor such as a central processing unit (CPU), or a processor dedicated to a specific purpose such as a graphics processing unit (GPU). Furthermore, explicit use of the terms "processor" or "controller" should not be construed to refer only to hardware capable of executing software; it may implicitly include, but is not limited to, digital signal processor (DSP) hardware, network processors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), read-only memory (ROM) for storing software, random access memory (RAM), and non-volatile storage. Other hardware, conventional and / or custom, may also be included.

[0080] Software modules, or modules that are simply implied to be software, may be represented herein as any combination of flowchart elements or other elements showing the execution of process steps and / or textual descriptions. Such modules may be executed by explicitly or implicitly shown hardware.

[0081] With these fundamentals in mind, we now consider some non-limiting implementations of the present technology.

[0082] Referring to FIG. 1, a schematic diagram of a computing device 100 suitable for use in some non-limiting implementations of the present technology is shown.

[0083] Computing Devices Computing device 100 comprises various hardware components, including one or more single-core or multi-core processors, collectively referred to as processor 110, a graphics processing unit (GPU) 111, a solid-state drive 120, random access memory 130, a display interface 140, and an input / output interface 150.

[0084] Communication between the various components of computing device 100 may be enabled by one or more internal and / or external buses 160 (e.g., PCI bus, Universal Serial Bus, IEEE 1394 "Firewire" bus, SCSI bus, Serial ATA bus, etc.) to which the various hardware components are electronically coupled.

[0085] Input / output interface 150 may be coupled to touchscreen 190 and / or to one or more internal and / or external buses 160. Touchscreen 190 may be part of a display. In some implementations, touchscreen 190 is a display. Touchscreen 190 may also be referred to as screen 190. In the implementation shown in FIG. 2, touchscreen 190 includes touch hardware 194 (e.g., pressure-sensitive cells embedded in a layer of the display that enable detection of physical interaction between a user and the display) and a touch input / output controller 192 that enables communication with display interface 140 and / or one or more internal and / or external buses 160. In some implementations, input / output interface 150 may be connected to a keyboard (not shown), a mouse (not shown), or a trackpad (not shown), allowing a user to interact with computing device 100 in addition to or instead of touchscreen 190.

[0086] According to an embodiment of the present technology, the solid-state drive 120 stores program instructions suitable for being loaded into the random access memory 130 and executed by the processor 110 and / or GPU 111 to generate a parametric mesh. For example, the program instructions may be part of a library or an application.

[0087] Computing device 100 may be implemented in the form of a server, a desktop computer, a laptop computer, a tablet, a smartphone, a personal digital assistant, or any device that can be configured to implement the present technology, as will be understood by those skilled in the art.

[0088] system Referring to FIG. 2 , a schematic diagram of a communication system 200 is shown, referred to as system 200, suitable for implementing a non-limiting embodiment of the present technology. It should be clearly understood that the illustrated system 200 is merely an exemplary embodiment of the present technology. Accordingly, the following description thereof is intended only as a description of an exemplary example of the present technology. This description is not intended to define the scope or delimit the present technology. In some cases, what are believed to be useful examples of modifications to system 200 are described below. This is done merely to facilitate understanding and, again, is not intended to define the scope or delimit the present technology. These modifications are not an exhaustive list, and other modifications are likely possible, as would be understood by one skilled in the art. Furthermore, if a modification is not made (i.e., if no example of a modification is mentioned), it should not be interpreted as meaning that the modification is not possible and / or that what is described is the only way to implement that element of the present technology. As would be understood by one skilled in the art, this is likely not the case. Furthermore, it will be understood that system 200 may, in certain instances, provide a simple implementation of the present technology, and in such cases, they are presented in this manner to aid in understanding. As will be appreciated by those skilled in the art, various implementations of the present technology may be more complex.

[0089] The system 200 includes, among other things, one or more medical imaging devices 210, a server 230, and a database 235 coupled via a communications network 220 by individual communications links 225 (not separately numbered).

[0090] In one or more embodiments, at least a portion of the system 200 implements Picture Archiving and Communication System (PACS) technology.

[0091] One or more medical imaging devices 210 are operated by a user (eg, a doctor or technician) to acquire medical images of a given patient's body.

[0092] Medical Imaging Equipment The one or more medical imaging devices 210 are referred to herein as medical imaging devices 210 .

[0093] The medical imaging device 210 is configured, among other things, to (i) acquire one or more images of an anatomical structure of interest of a given patient according to acquisition parameters, and (ii) transmit the images to the workstation computer 215 and / or the server 230.

[0094] The medical imaging device 210 may comprise one of an X-ray device, a computed tomography (CT) scanner, a magnetic resonance imaging (MRI) scanner, ultrasound, or the like.

[0095] In some implementations of the present technology, the medical imaging device 210 may include multiple medical imaging devices such as, but not limited to, an X-ray device, a computed tomography (CT) scanner, a magnetic resonance imaging (MRI) scanner, ultrasound (including 2D or 3D ultrasound), positron emission tomography (PET), single photon emission computed tomography (SPECT), and the like.

[0096] The medical imaging device 210 may be configured with specific acquisition parameters for acquiring images of a patient including one or more anatomical structures of interest.

[0097] As a non-limiting example, in one or more embodiments in which the medical imaging device 210 is implemented as a CT scanner, a CT protocol may be used that includes a preoperative retrospective gated multi-detector CT (MDCT—64-row multi-slice CT scanner) that may use variable dose radiation to capture RR intervals.

[0098] As another non-limiting example, in one or more embodiments in which the medical imaging procedure includes an MRI scanner, the MR protocol can include steady-state T2-weighted fast field echo (TE=2.6 ms, TR=5.2 ms, flip angle 110 degrees, fat suppression (SPIR), echo time 50 ms, maximum 25 cardiac phases, matrix 256x256, acquisition voxel MPS (measurement, phase, and slice encoding directions) 1.56 / 1.56 / 3.00 mm, and reconstruction voxel MPS 0.78.

[0099] In one or more alternative embodiments, the medical imaging device 210 may include or be connected to a workstation computer 215 for controlling, among other things, acquisition parameters and transmission of image data.

[0100] Workstation Computer The workstation computer 215 is configured to, among other things, (i) control acquisition parameters of the medical imaging device to perform medical imaging, (ii) receive and process images from the medical imaging device 210, and (iii) transmit images to the server 230.

[0101] The workstation computer 215 is configured to control acquisition parameters of the medical imaging device 210 .

[0102] The workstation computer 215 may receive images in raw format from the medical imaging device 210 and perform tomographic reconstruction using known algorithms and software.

[0103] Implementations of workstation computer 215 are known in the art. Workstation computer 215 may be implemented as computing device 100 or as components of computing device 100, such as processor 110, graphics processing unit (GPU) 111, solid-state drive 120, random access memory 130, display interface 140, and input / output interface 150.

[0104] In one embodiment, the workstation computer 215 is configured according to the Digital Imaging and Communications in Medicine (DICOM) standard for communication and management of medical imaging information and related data.

[0105] The workstation computer 215 is connected to a server 230 via a communications network 220 by a communications link (not numbered).

[0106] In one or more alternative embodiments, the workstation computer 215 may be provided along with the medical imaging device 210. In one or more other embodiments, the workstation computer 215 may be implemented as a mobile device such as a smartphone or tablet.

[0107] In one or more embodiments, the medical imaging device 210, along with other electronic devices such as a server 230, is part of a Picture Archiving and Communication System (PACS) for communicating and managing medical imaging information and related data.

[0108] server The server 230 is configured to, among other things, (i) initialize a mesh of one or more anatomical structures according to a set of mesh parameters; (ii) receive an image set of a given patient including at least a portion of one or more anatomical structures, the image set being acquired by the medical imaging device 210; (iii) segment the image set to obtain a set of segmented images including a plurality of anatomical segments; (iv) receive a plurality of domain representations of one or more anatomical structures including individual biomarkers; (v) determine individual correspondence rules between each domain representation and the initial mesh; and (vi) encode the initial mesh with each individual feature set obtained from the biomarkers in the domain representation based on the individual correspondence rules to obtain a parametric mesh, wherein each node of the parametric mesh includes multiple feature channels including the individual feature sets.

[0109] In some implementations, the server 230 may have access to a set of machine learning models 250 to perform some of the processes described above.

[0110] How server 230 is configured is described in more detail later in this specification.

[0111] Server 230 may be implemented as a conventional computer server and may include some or all of the components of computing device 100 shown in FIG. 2. In one or more example implementations of the present technology, server 230 may be implemented as a Dell™ PowerEdge™ server running the Microsoft™ Windows Server™ operating system. Of course, server 230 may be implemented with any other suitable hardware and / or software and / or firmware, or combination thereof. In the illustrated non-limiting embodiment of the present technology, server 230 is a single server. In alternative non-limiting implementations of the present technology, the functionality of server 230 may be distributed and implemented across multiple servers (not shown).

[0112] Implementations of server 230 are well known to those skilled in the art. However, briefly described, server 230 comprises a communication interface (not shown) structured and configured to communicate with various entities (e.g., workstation computer 215 and other devices that may be connected to network 220) over communication network 220. Server 230 further comprises at least one computer processor (e.g., processor 110 or GPU 111 of computing device 100) operably coupled to the communication interface and structured and configured to execute the various processes described herein.

[0113] In one or more embodiments, server 230 may be implemented as computing device 100 or may include components of computing device 100 such as processor 110, graphics processing unit (GPU) 111, solid-state drive 120, random access memory 130, display interface 140, and input / output interface 150.

[0114] It will be appreciated that server 230 can provide the output of one or more processing steps to another electronic device for display, review, and / or problem resolution. As a non-limiting example, server 230 can transmit images, calculations, results, machine learning parameters for display on a client device configured similarly to computing device 100, such as a smartphone, tablet, or the like.

[0115] The server 230 has access to a set of machine learning (ML) models 250 .

[0116] Machine Learning (ML) Models The set of ML models 250 includes, among other things, a set of segmentation ML models 260 and a set of growth prediction ML models 270.

[0117] The ML model will be referred to hereafter as the model.

[0118] Each of the set of models 250 is parameterized by, among other things, model parameters and hyperparameters.

[0119] Model parameters are the configuration variables of the model used to perform predictions and are estimated or learned from training data, i.e., the coefficients are selected during training based on an optimization strategy to output predictions. Hyperparameters are the configuration variables of the model that determine the structure of the initial model and how it is trained.

[0120] It will be appreciated that the number of model parameters to initialize will depend, among other things, on the type of model (e.g., classification model or regression model), the model architecture (e.g., DNN, SVM, ensemble tree, etc.), and the model's hyperparameters (e.g., number of layers, type of layers, number of neurons in the NN).

[0121] In one or more embodiments, the hyperparameters include one or more of the number of hidden layers and units, the optimization algorithm, the learning rate, the momentum, the activation function, the mini-batch size, the number of epochs, and dropout.

[0122] Segmentation Model The set of segmentation models 260 includes one or more segmentation models.

[0123] The set of segmentation models 260 is configured to perform segmentation of anatomical structures in images acquired by a medical imaging modality, such as medical imaging device 210 .

[0124] In one or more embodiments, the set of segmentation models 260 is configured to detect (i.e., define) all boundaries within an image, including anatomical structures, and to identify (i.e., classify) various tissue types.

[0125] In one or more embodiments, when the anatomical structure of interest includes an aortic region, the set of segmentation models 260 is configured to segment the outer wall of the aorta, the inner wall of the aorta, the lumen, and the intraluminal thrombus (ILT). Thus, the segmentation model 260 can classify each pixel in the image as one of the outer wall of the aorta, the inner wall of the aorta, the lumen, the intraluminal thrombus (ILT), and the background.

[0126] In one or more embodiments, the set of segmentation models 260 refers to multiple segmentation models 260, each configured to perform a particular segmentation task. As a non-limiting example, the segmentation models 260 may include a first segmentation model configured to perform foreground and background segmentation, a second segmentation model configured to perform semantic segmentation of the lumen within the aorta, and a third model configured to perform pathological tissue classification (e.g., classification of calcified and non-calcified tissue within the aortic wall and intraluminal thrombus, if present). Non-limiting examples of such segmentation models are described in International Patent Application No. PCT / IB2022 / 051558, entitled "METHOD AND SYSTEM FOR SEGMENTING AND CHARACTERIZING AORTIC TISSUES," filed February 22, 2022 by the same applicant, the contents of which are incorporated herein by reference.

[0127] In one or more implementations, the set of segmentation models 260 may be composed of convolutional neural network layers (e.g., U-Net or V-Net based), attention-based mechanisms (i.e., transformer-based models such as Vision Transformer (ViT) models), and combinations thereof. It will be appreciated that the set of segmentation models 260 may use an encoder-decoder architecture.

[0128] In one or more embodiments, the set of segmentation models 260 may be based on fully convolutional neural networks (FCNs), generative adversarial networks (GANs), cascade networks, etc.

[0129] In one or more other implementations, the segmentation model 260 may have a ResNet-based FCN architecture. Non-limiting examples of ResNets include ResNet50 (50 layers), ResNet101 (101 layers), ResNet152 (152 layers), ResNet50V2 (50 layers with batch normalization), ResNet101V2 (101 layers with batch normalization), and ResNet152V2 (152 layers with batch normalization).

[0130] In one or more alternative embodiments, the set of segmentation models 260 may be implemented based on one of U-Net, V-Net, SegNet, AlexNet, GoogleNet, VGG, DeepLab, Mask R-CNN, etc.

[0131] Growth Prediction Model The set of growth forecast models 270 includes one or more growth forecast models 270.

[0132] The set of growth prediction models 270 is configured to, among other things, (i) receive one or more images of the patient's anatomical structures, including the aorta; (ii) generate or extract one or more of topographical features and functional features; and (iii) use the features to perform an individualized prediction indicative of AAA growth.

[0133] In some embodiments, the set of growth predictive ML models 270 is configured to (i) obtain a parametric mesh of the patient encoding structural and functional features generated based on images of the patient, and (ii) generate a prediction indicative of AAA growth based on the selected features.

[0134] In one or more embodiments, the set of growth predictive ML models 270 is configured to generate functional and regional characterizations of aortic tissue, including intraluminal thrombus (ILT) thickness and wall shear stress, and generate predictions indicative of AAA growth.

[0135] In one or more alternative embodiments, the set of growth predictive ML models 270 is configured to generate functional and regional characterizations of aortic tissue, including strain, intraluminal thrombus (ILT) thickness, and wall shear stress, and generate predictions indicative of AAA growth.

[0136] In the context of the present technology, the individual prediction is indicative of AAA growth. In one or more embodiments, the individual prediction can be either non-significant AAA growth or significant AAA growth. In one or more other embodiments, the individual prediction is a multi-class prediction indicative of AAA growth.

[0137] In one or more embodiments, the set of growth prediction models 270 includes a plurality of classification models, which may be divided into subsets of classification models, each of which may be configured to perform predictions based on different types of features, as described below.

[0138] As non-limiting examples, the set of growth prediction models 270 may use ensemble trees, support vector machines (SVMs), random forests, neural networks, and the like.

[0139] In some embodiments, the set of growth prediction models 270 is implemented using extra trees.

[0140] Database The database 235 is configured to, among other things, (i) store acquisition parameters and data related to the medical imaging device 210, (ii) store images acquired by a medical imaging modality such as the medical imaging device 210, (iii) store data related to a set of ML models 250, including model parameters, hyperparameters, datasets, and outputs, (iv) store data related to meshes, (v) store multiple domain representations including biomarkers, and (vi) generate parametric meshes including all their feature channels.

[0141] Database 235 is configured to store images and videos. In one or more embodiments, the database can store Digital Imaging and Communications in Medicine (DICOM) files, including, for example, DCM and DCM30 (DICOM 3.0) file extensions. Additionally or alternatively, database 235 can store medical image files in Tag Image File Format (TIFF), Digital Storage and Retrieval (DSR) TIFF-based format, and Data Exchange File Format (DEFF) TIFF-based format.

[0142] In one or more embodiments, database 235 may also store ML file formats such as .tfrecords, .csv, .npy, and .petastorm, as well as file formats used to store models such as .pb, .pkl, .pt, or .pth. Database 235 may also store known file formats such as image file formats (e.g., .png, .jpeg, .exif, .bmp, .tiff), video file formats (e.g., .mp4, .mkv, etc.), archive file formats (e.g., .zip, .gz, .tar, .bzip2), document file formats (e.g., .docx, .pdf, .txt), or web file formats (e.g., .html).

[0143] It will be appreciated that the database 235 may store other types of data, such as a validation data set (not shown), a test data set (not shown), and the like.

[0144] communication network In some implementations of the present technology, the communications network 220 is the Internet. In alternative non-limiting implementations, the communications network 220 may be implemented as any suitable local area network (LAN), wide area network (WAN), private communications network, etc. It should be expressly understood that the implementation of the communications network 220 is for illustrative purposes only. How the communications links 225 (not separately numbered) between the workstation computer 215 and / or server 230 and / or another electronic device (not shown) and the communications network 220 is implemented will depend, among other things, on how each of the medical imaging device 210, the workstation computer 215, and the server 230 is implemented.

[0145] Communications network 220 may be used to transmit data packets between workstation computer 215, server 230, and database 235. For example, communications network 220 may be used to transmit requests between workstation computer 215 and server 230.

[0146] Parametric mesh generation procedure Referring to FIG. 3, a schematic diagram of a parametric mesh generation procedure 300 is shown in accordance with one or more non-limiting embodiments of the present technique.

[0147] The goal of the parametric mesh generation procedure 300 is to generate a three-dimensional (3D) parametric mesh that is a structural representation of one or more anatomical structures of a given patient, where each element or node of the parametric mesh encodes, among other things, structural, temporal, functional, and other descriptive information (also called biomarkers) at the corresponding location, which may be acquired using different imaging modalities and / or physical domain representations. How the parametric mesh generation procedure 300 is configured to achieve that goal is described in more detail below. How the parametric mesh generation procedure 300 is configured to achieve that goal is described in more detail below.

[0148] The parametric mesh generation procedure 300 includes, among other steps, a mesh initialization procedure 320, an image acquisition procedure 330, a segmentation procedure 340, a multi-domain data acquisition procedure 350, a registration procedure 380, and a parametric mesh encoding procedure 390.

[0149] In one or more embodiments, the parametric mesh generation procedure 300 may be performed by the server 230. In one or more alternative embodiments, the parametric mesh generation procedure 300 may be performed by one or more computing devices in a distributed manner. As a non-limiting example, a first computing device, such as the server 230, may perform at least a portion of the parametric mesh generation procedure 300 (i.e., one of the steps), and one or more other computing devices may perform other portions of the parametric mesh generation procedure 300 (i.e., other of the steps).

[0150] The parametric mesh generation procedure 300 includes a mesh initialization procedure 320 .

[0151] Mesh initialization procedure The mesh initialization procedure 320 is configured to, among other things, (i) receive a mesh parameter set and (ii) generate an initial 3D mesh based on the mesh parameter set. The mesh initialization procedure 320 allows for the specification of mesh dimensions, geometry, and other attributes (e.g., initial conditions or parameters) before the mesh is loaded with patient-specific biomarker data. The mesh initialization procedure 320 serves as a preprocessing step to facilitate subsequent encoding of the mesh with biomarker data for analysis or visualization.

[0152] In the context of the present technology, the mesh initialization procedure 320 is used to initialize a 3D mesh, which is a multidimensional array that provides, among other things, a spatial representation of one or more anatomical structures in the form of a 3D geometric shape including a discrete number of volume elements, also called "elements" or "cells." It will be appreciated that the mesh may provide a discrete spatial and temporal representation of one or more anatomical structures for a given patient. The nodes of the initial 3D mesh are then populated or encoded with single-modal or multi-modal data, i.e., one or more biomarkers, for the same patient, resulting in a 3D parametric mesh, as described below. The 3D parametric mesh may be used to store biomarker data from multiple domains in the form of features and to visually render the features in time.

[0153] The mesh initialization procedure 320 may be performed any time before the registration procedure 380 .

[0154] The mesh initialization procedure 320 receives a set of mesh parameters. The set of mesh parameters may be entered by an operator of the technology via an input / output device, such as a keyboard, touch screen, etc. In one or more embodiments, the set of mesh parameters may be received from another client device.

[0155] The set of mesh parameters may specify the configuration of the mesh for each anatomical structure to be represented and may include one or more of its geometry, number of nodes, type of volume elements, and number of volume elements.

[0156] In one or more embodiments, the one or more anatomical structures represented by the mesh include the aorta. Additionally, the one or more anatomical structures represented by the mesh may include the iliac arteries.

[0157] The mesh parameter set includes a predetermined total number of nodes that form the initial 3D mesh. It will be understood that the number of nodes is not limited and may include, as a non-limiting example, 10,000 nodes. In one or more alternative implementations, a number of predetermined feature channels may be associated with each node.

[0158] The total number of nodes may be set by an operator of the present technology. In some alternative implementations of the present technology, the number of nodes may depend on the number of feature channels, the computational resources (e.g., the storage capacity of the electronic device used to implement the present technology), and the intended use of the 3D parametric mesh.

[0159] In one or more embodiments, the total number of nodes in the 3D mesh and the number of nodes in at least one of each concentric 3D mesh layer, each layer parallel to the axial plane, and each layer parallel to the sagittal plane can also be predetermined.

[0160] The initial 3D mesh includes or defines multiple concentric 3D mesh layers positioned relative to a centerline. Each concentric 3D mesh layer may be understood to be a separate 3D mesh with its nodes positioned at a separate distance from a separate section of a shared centerline (e.g., a 3D point or line defining the center of the represented anatomical structure). Briefly referring to Figures 5A and 5B, Figure 5B shows the first parametric mesh 400 with the top portion removed to reveal multiple concentric 3D mesh layers 440, which are described in more detail below.

[0161] Returning to Figure 3, multiple concentric 3D mesh layers can be used to represent different boundaries of the anatomical structures, substructures, and / or spaces between them for which information is to be encoded. By way of non-limiting example, the initial 3D mesh can include 3D mesh layers referenced to an outer mesh representing the outer surface of the anatomical structure and / or referenced to an innermost mesh at the centerline of the anatomical structure. It will be appreciated that in one or more alternative embodiments, the 3D mesh can include one or more additional layers outside the anatomical structure.

[0162] The centerline of the 3D mesh may correspond to the centerline of the anatomical structure of interest, as determined during the registration procedure 380. It should be understood that sections of the centerline may extend vertically and horizontally in 3D, and that the centerline generally follows the shape of the anatomical structure being represented.

[0163] For a given fixed ordinate (i.e., a fixed value on the vertical or z-axis) corresponding to an axial (transverse) slice (i.e., parallel to a transverse or axial plane), each concentric 3D mesh layer is sometimes referred to as a two-dimensional (2D) concentric mesh axial layer or 2D mesh axial slice. Referring briefly to Figure 6, an axial view of the parametric mesh 400 is shown showing multiple 2D concentric mesh layers 440, which will be described in more detail herein below.

[0164] Returning to FIG. 3 , it will be appreciated that when the initial 3D mesh is initialized, its 3D visual representation has not yet been generated. This is because the initial 3D mesh has not yet been encoded and populated with data from actual measurements of the patient's anatomical structure(s). Thus, the initial 3D mesh is "generic" or the same for each patient before being encoded with patient-specific data and, optionally, visually represented on a graphical user interface. In one or more alternative embodiments, the initial 3D mesh may be represented in 3D using a generic or "default" shape of the anatomical structure.

[0165] Each concentric 3D mesh layer may be "unwrapped" and represented as a 2D array, with each element in the 2D array corresponding to a distinct node in the respective concentric 3D mesh layer. Each node is associated with or represented by a distinct node array, sometimes referred to as a "feature channel," that encodes node features. Each location in the node array corresponds to a distinct feature that will be encoded in a distinct node of the 3D mesh upon completion of the parametric mesh generation procedure 300.

[0166] Each node of the initial 3D mesh is associated with a distinct node position coordinate. It will be appreciated that the node position coordinates may be expressed in different and equivalent ways depending on the coordinate system used. The node position coordinates are used to reference the nodes of the mesh or associated array and to store and retrieve information.

[0167] In one or more embodiments, a given node may be identified based on the concentric 3D mesh layer in which it is located using a mesh layer coordinate R. The mesh layer coordinate R may be used to identify the particular concentric layer array in which the node information is stored, as each concentric layer array corresponds to a different array of the same size (i.e., the same number of nodes).

[0168] In one or more embodiments, each node may be identified for each concentric 3D mesh layer, and the coordinate M may refer to the node's circumferential position on the concentric 3D mesh layer (i.e., corresponding to its position along the circumference on an axial slice of the 2D mesh), and the coordinate N may refer to the node's longitudinal position (i.e., corresponding to the 2D axial slice to which the node belongs). Each concentric 3D mesh layer may be "unwrapped" and correspond to a 2D array, so that the node coordinates in the 2D array are the same as the node coordinates in the 3D mesh.

[0169] In some embodiments, each node of the 3D mesh is associated with an individual node identifier, which may include one or more numbers to uniquely identify the node for purposes of retrieval and / or storage of information (e.g., in multiple feature channels). It will be understood that the node coordinates are the same for each patient.

[0170] In one or more embodiments, each node is associated with a distinct node position coordinate and a distinct node time coordinate. The node time coordinate may include two or more time coordinates that allow the node to be represented at different discrete points in time. It will be appreciated that in some embodiments of the present technology, a time coordinate (i.e., a time frame) may not be required. Thus, using time coordinates, the 3D mesh can visually represent changes over time (e.g., changes in the geometry of blood vessels during a cardiac cycle).

[0171] In one or more embodiments, each concentric 3D mesh layer can include MxNxP nodes, where M is the number of circumferential points, N is the number of longitudinal points, and P is the number of time frames. It will be understood that a planar section (i.e., a layer) of the mesh corresponds to all nodes of fixed ordinate coordinate N and fixed temporal coordinate O.

[0172] It will be appreciated that a 3D mesh can be represented as a collection of 2D arrays, with each concentric 3D mesh layer corresponding to a respective 2D array of the same size (i.e., the same number of nodes), with each 2D array element corresponding to a node and including a respective array corresponding to the node's feature channel. In one or more embodiments, a time frame may also be encoded in the collection of 2D arrays, in this case, a given collection of 2D arrays (corresponding to the complete 3D mesh). In other embodiments, a time value may be encoded at each node (i.e., each node may correspond to an array with a feature channel for each instant).

[0173] It is important to understand that each node of a patient's 3D mesh encodes the patient's actual anatomical geometry information into the node's feature channels, and the patient-specific 3D mesh is visually rendered using the feature channels that encode the information.

[0174] The mesh initialization procedure 320 outputs an initial 3D mesh.

[0175] In one or more embodiments, the initial 3D mesh is represented as a multidimensional array including a separate concentric layer array representing each concentric 3D mesh layer, the size of each separate concentric layer array corresponding to the number of nodes in the separate concentric 3D mesh layer, and each node including a separate node array for storing node characteristics, each separate node array capable of storing node characteristics at discrete moments in time.

[0176] Image acquisition procedure The image acquisition procedure 330 is configured, among other things, to receive a set of images of the patient's body acquired by the medical imaging device 210 .

[0177] The set of images of the patient's body includes at least one image of the patient's body that is a discrete representation of a signal containing at least a portion of one or more anatomical structures of interest, generated using the medical imaging device 210.

[0178] It will be appreciated that in the case of a 2D domain representation, the image cells are called "pixels" and in the case of a 3D domain representation, the image cells are called "voxels."

[0179] In one or more embodiments, the set of images may be in the form of an image stack.

[0180] It will be appreciated that an image stack comprises a set of consecutive images, also called scrollable slices, as would be expected for cross-sectional studies (e.g., CT / MRI) or time-resolved modalities. As a non-limiting example, an image stack may be provided in DICOM file format.

[0181] In one or more embodiments, the image stack may be in the form of a multi-phase stack, and each phase of the multi-phase stack may correspond to a time instance. As a non-limiting example, each phase in the stack may correspond to a moment in a given patient's cardiac cycle.

[0182] In one or more embodiments, the set of images of the patient's body includes the aorta(s) and / or the iliac arteries.

[0183] In one or more embodiments, one or more anatomical structures in the image set may include the thoracic region (e.g., ascending aorta, aortic arch, descending thoracic aorta) and / or abdominal aortic region (e.g., suprarenal abdominal aorta, infrarenal aorta, renal arteries, lumbar arteries) and iliac arteries (e.g., common iliac arteries, external iliac arteries, internal iliac arteries).

[0184] It will be appreciated that the image acquisition procedure 330 may receive multiple image sets of a given patient's body, each image set corresponding to a different imaging session. It will be appreciated that a given image set may be selected as a "baseline" set, and image sets acquired during other imaging sessions may be sent to the multi-domain data acquisition procedure 350 and subsequently encoded.

[0185] In one or more other embodiments, the image acquisition procedure 330 can receive multiple image sets, one or more of the image sets being acquired using different medical imaging modalities.

[0186] The image acquisition procedure 330 outputs a set of images.

[0187] Segmentation Procedure The segmentation procedure 340 is configured to, among other things, (i) receive an image set; and (ii) segment the image set to obtain a set of segmented images including a plurality of anatomical segments.

[0188] The segmentation procedure 340 uses manual and / or automatic segmentation methods to obtain a plurality of anatomical segments, which may also be referred to as segmented tissues.

[0189] In one or more other embodiments, the segmentation procedure 340 may use manual segmentation techniques to obtain the segmented image.

[0190] In one or more embodiments, the segmentation procedure 340 obtains one or more anatomical segments using a set of trained segmentation ML models 260 that are trained to segment anatomical structures in images acquired by an imaging device (e.g., medical imaging device 210). For example, the segmentation procedure 340 may output one of a number of classes for a region in the image, including at least one anatomical segment and background.

[0191] As a non-limiting example, the set of segmentation models 260 may be trained to segment aortic tissue in an image.

[0192] In one or more embodiments, the segmentation procedure 340 obtains a segmented aortic region that includes one or more of the aorta and iliac arteries for each image in the image set. In one or more embodiments, the segmented aortic region includes an ROI that includes the lumen, the aortic wall, the ILT (if present), and calcification (if present).

[0193] Additionally, the segmentation procedure 340 can segment the different branches of the abdominal aorta, including the celiac trunk, the superior and inferior mesenteric arteries, the hepatic artery, the splenic artery, the renal arteries, and the iliac arteries.

[0194] In one or more alternative embodiments, the segmentation procedure 340 outputs a set of segmented images, with each pixel labeled with a distinct segmented tissue label.

[0195] In one or more embodiments, the segmentation procedure 340 extracts segmented tissue from the set of images to obtain at least one image per anatomical segment. It will be appreciated that the segmented tissue can be extracted by performing masking.

[0196] In one or more embodiments, the segmentation procedure 340 determines a center point for each of the anatomical segments, which can be used by the registration procedure 380 to determine the 3D centerline of each anatomical segment.

[0197] The segmentation procedure 340 outputs a representation of multiple anatomical segments for each of the set of images.

[0198] Multi-domain data acquisition procedure The multi-domain data acquisition procedure 350 is configured to, among other things, (i) transmit an image set and / or a segmented image set including multiple anatomical segments, and (ii) receive one or more domain representations including biomarkers associated with anatomical structures within the body.

[0199] In one or more embodiments, the multi-domain data acquisition procedure 350 transmits a set of images and / or a set of segmented images so that different domain representations can be generated based on the transmitted set of images and / or multiple segments, each domain representation including biomarkers associated with an anatomical segment.

[0200] In the context of the present technology, a domain representation should be understood to be a discretized 2D and / or 3D representation of one or more anatomical structures of interest of a given patient, which may or may not include a discretized temporal representation. The domain representation may be computed and / or obtained using one or more imaging modalities.

[0201] The domain representation may include or be associated with biomarker values ​​that may provide structural, functional, and temporal information for elements contained in one or more anatomical structures in the domain representation. Non-limiting examples of biomarkers include pixel intensity, pixel location, structural mechanics values ​​(e.g., pressure, strain, stress, etc.), flow-related values ​​(e.g., velocity), and any other descriptive variables.

[0202] Each domain representation may have a different data format and / or data density and / or resolution.

[0203] In one or more embodiments, the given domain representation may be a mesh, such as a polygonal mesh. A polygonal mesh includes vertices, edges, and faces. The faces may include one of triangles (triangle mesh), quadrilaterals (quads), convex polygons (n-gons), concave polygons, and polygons with holes. The mesh may be a 2D or 3D mesh, with or without time discretization.

[0204] In one or more other embodiments, the given domain representation may be a 2D or 3D image.

[0205] In one or more embodiments, the multi-domain data acquisition procedure 350 sends the set of images and / or the set of segmented images to the fluid dynamics procedure 360 ​​.

[0206] Fluid Dynamics Procedure The fluid dynamics procedure 360 ​​is configured to perform a fluid dynamics analysis to generate a fluid dynamics representation that includes fluid dynamics biomarkers.

[0207] It will be appreciated that the fluid dynamics procedure 360 ​​may be performed by one or more other electronic devices and / or the server 230 .

[0208] In one or more alternative implementations of the present technology, the fluid dynamic representation including the fluid dynamic biomarkers may be generated or extracted by performing a dynamic imaging method (eg, 4D flow MRI).

[0209] In one or more embodiments, the fluid dynamics procedure 360 ​​simulates complex flows through a numerical discretization and solution approach using computational fluid dynamics (CFD) to obtain numerical solutions at discrete time / space points in the flow field. In one or more embodiments, the fluid dynamics biomarkers may include variables related to heat transfer or fluid flow.

[0210] In one or more embodiments, the fluid dynamics procedure 360 ​​performs spatial discretization or meshing based on the set of segmented images to divide the geometry into several discrete volume elements or cells, and then performs time discretization. The fluid dynamics procedure 360 ​​establishes boundary conditions, i.e., a set of applied physiological parameters (which may vary over time) that define the physical conditions at the inlets, outlets, and walls. It will be understood that the boundary conditions may be based on patient-specific data, population data, physical models, or assumptions. Further properties for the simulation are defined, including blood density and viscosity (i.e., fluid model), initial conditions of the system (e.g., whether the fluid is initially static or moving), time discretization information (time step size and numerical approximation scheme), and / or desired output data (e.g., number of cardiac cycles to simulate). The fluid dynamics procedure 360 ​​uses a CFD solver to solve the Navier-Stokes and continuity equations, incrementally progressing toward convergence to obtain a final solution. Next, the fluid dynamics procedure 360 ​​obtains fluid dynamics biomarkers including the pressure and velocity fields across all elements at each time step. It will be appreciated that additional biomarkers may be calculated and obtained based on the foregoing.

[0211] In one or more embodiments, the fluid dynamics procedure 360 ​​may use volumetric meshes ranging from 2 million to 3.5 million elements in different possible geometries.

[0212] As a non-limiting example, the fluid mechanics procedure 360 ​​performs a sensitivity analysis to obtain a volumetric mesh of approximately 2 million tetrahedral elements, and runs a CFD simulation in ANSYS™ software FLUENT using a semi-implicit solution of the pressure-coupled equations (SIMPLE) and a second-order implicit transient formulation, assuming laminar blood flow, a time-varying velocity profile based on the flow rate of the descending aorta at the inlet of the fluid domain, and an outflow boundary condition of 50% flow split to the iliac arteries. The rheology model can assume that blood is an isotropic, incompressible, Newtonian fluid assigned a constant density (1060 kg / m3) and dynamic viscosity (0.00319 Pa s). The arterial wall can be assumed to be rigid, and a no-slip condition can be applied at the fluid interface. The fluid mechanics procedure 360 ​​can output fluid dynamics biomarkers for elements at the boundary of the computational domain. In the aforementioned non-limiting example, the fluid dynamics procedure 360 ​​may output biomarkers such as blood velocity and wall shear stress at the boundary of a mesh containing two tetrahedral million elements.

[0213] As a non-limiting example, for the aorta, models such as Windkessel, a distributed model of arterial behavior, a reservoir pressure model, or finite element analysis can be used to obtain fluid dynamic biomarkers.

[0214] The fluid dynamics procedure 360 ​​outputs a fluid dynamics representation that includes the fluid dynamics biomarkers.

[0215] The fluid dynamic representation specifies information related to the spatial and temporal discretization, and the fluid dynamic biomarker specifies several variables and their values ​​for each of the elements in the spatial and temporal discretization.

[0216] The multi-domain data acquisition procedure 350 receives a fluid dynamics representation that includes fluid dynamics biomarkers.

[0217] In one or more embodiments, the multi-domain data acquisition procedure 350 sends the set of images and / or the set of segmented images to the structural mechanics procedure 365 .

[0218] Structural Mechanics Procedures The structural mechanics procedure 365 is configured to perform a structural mechanics analysis to generate a structural mechanics representation that includes structural mechanics biomarkers, which are variables that indicate the structural and mechanical properties of one or more anatomical structures of interest.

[0219] It will be appreciated that the structural mechanics procedure 365 may be performed by one or more other electronic devices and / or the server 230 .

[0220] The structural mechanics procedure 365 uses a set of segmented images containing multiple anatomical segments to generate a structural mechanics representation, which is used to obtain structural mechanics biomarkers.

[0221] The structural mechanics procedure 365 can obtain structural mechanics biomarkers using domain representations that are different from other domain representations (eg, fluid mechanics and descriptive variable representations).

[0222] In one or more embodiments, the structural mechanics procedure 365 performs spatial discretization or meshing based on the set of segmented images to divide the geometry into a number of discrete surface elements or cells, and then performs temporal discretization. The structural mechanics procedure 365 may generate a specific mesh having tetrahedral, triangular, hexahedral, triangular, and / or rectangular elements to calculate unimodal or multimodal structural mechanics biomarkers. The structural mechanics procedure 365 then performs mechanical property characterization based on the generated mesh. It will be appreciated that various methods, including finite element analysis simulation, can be used.

[0223] The structural mechanics representation is used to obtain structural mechanics biomarkers, which may include stress-strain relationships and strength of one or more anatomical structures of interest.

[0224] Structural mechanics biomarkers may include thickness, strain, and stress, as well as other biomarkers derived based on the foregoing.

[0225] In one or more embodiments, the structural mechanics procedure 365 acquires structural mechanics biomarkers including strain, which may further include maximum principal strain, minimum principal strain, circumferential strain, and longitudinal strain, corresponding strain rates, and / or timing of maximum strain.

[0226] As a non-limiting example, the structural mechanics procedure 365 may generate a surface wall mesh of approximately 4000 triangular shell elements and track nodal velocities in a three-dimensional image stack representing the aorta throughout the cardiac cycle. The structural mechanics procedure 365 may then measure nodal displacements based on the nodal velocities and calculate in vivo strains. In this non-limiting example, the structural mechanics procedure 365 uses a domain representation that includes a surface wall mesh of approximately 4000 triangular shell elements, which differs from the initial mesh used by the fluid mechanics procedure 360 ​​and the volumetric mesh of approximately 2 million tetrahedral elements.

[0227] Thus, in this non-limiting example, the structural mechanics procedure 365 uses a domain representation of the surface wall mesh of 4000 triangular elements and calculates strain biomarkers for the elements of the surface wall mesh.

[0228] As another non-limiting example, the structural mechanics procedure 365 can use the segmented aortic lumen and segmented wall received from the segmentation procedure 340 to generate an aortic lumen surface mesh and an aortic wall surface mesh, and calculate the ILT thickness by measuring the average distance between each mesh point of the outer wall mesh and an adjacent point on the lumen surface within a specified radius.

[0229] Non-limiting examples of methods and systems for generating structural mechanics representations and biomarkers, including regional rupture potential (RRP) of a blood vessel (also known as regional aortic weakening (RAW)), are described in detail in International Patent Application Publication WO 2021 / 059243 A1, entitled "METHOD AND SYSTEM FOR DETERMINING REGIONAL RUPTURE POTENTIAL OF BLOOD VESSEL," filed on September 25, 2020, by the same applicant, the contents of which are incorporated herein by reference.

[0230] The structural mechanics representation specifies information related to the spatial and temporal discretization of an anatomical structure(s), and the structural mechanics biomarkers specify several variables and their values ​​for each of the elements in the spatial and temporal discretization.

[0231] In one or more embodiments, the structural mechanics procedure 365 outputs one or more structural mechanics representations, each associated with a distinct structural biomarker. Each of the one or more structural mechanics representations may have a different mesh geometry and distinct structural biomarkers, and therefore may have a different data density.

[0232] The multi-domain data acquisition procedure 350 receives a structural dynamics representation that includes structural dynamics biomarkers.

[0233] Descriptive variable representation procedure The multi-domain data acquisition procedure 350 sends a set of images and / or a segmented image containing multiple segments to a descriptive variable representation procedure 370 to obtain a representation that includes other biomarkers not described above.

[0234] In one or more alternative embodiments, the multi-domain data acquisition procedure 350 may not send images or segments to the descriptive variable representation procedure 370, but may instead acquire one or more descriptive variable representations that include biomarkers that are not based on images or segments. This may be the case, for example, when other domain representations of the same anatomical structure of a given patient are acquired using other types of imaging modalities.

[0235] It will be appreciated that the descriptive variable representation procedure 370 may be performed by one or more other electronic devices and / or the server 230 .

[0236] In one or more embodiments, the descriptive variable representation procedure 370 may receive images of the same patient acquired by one or more other medical imaging devices different from the medical imaging device 210 .

[0237] In one or more embodiments, the one or more other medical imaging devices may include micro-CT, ultrasound, magnetic resonance (MR) imaging, positron emission tomography (PET) imaging, confocal microscopy, focused ion beam scanning electron microscope (FIBSEM), and the like.

[0238] Thus, it will be appreciated that the descriptive variable representation procedure 370 acquires images of at least a portion of the same patient anatomy.

[0239] As a non-limiting example, the descriptive variable representation procedure 370 can receive ultrasound images having Doppler velocities that may have a different resolution and may have a different acquisition angle and view than the set of images acquired by the image acquisition procedure 310.

[0240] The multi-domain data acquisition procedure 350 receives one or more descriptive variable representations that include individual descriptive biomarkers.

[0241] In one or more embodiments, the parametric mesh generation procedure 300 includes an alignment procedure 380 .

[0242] Alignment Procedure The registration procedure 380 is configured to, among other things, (i) receive multiple domain representations from the multi-domain data acquisition procedure 350, (ii) align the multiple domain representations to a common reference frame, and (iii) calculate individual correspondence rules between the initial mesh and each of the multiple domain representations.

[0243] In the context of the present technology, the registration procedure 380 is used to bring the various relevant representations and modalities into a common frame of reference (i.e., spatial alignment) so that the information they contain can be optimally integrated into a parametric mesh during the parametric mesh encoding procedure 390.

[0244] In one or more embodiments, the registration procedure 380 is configured to receive results of multi-temporal image analysis where images of the same patient were acquired at different times and / or under different health conditions.

[0245] In one or more embodiments, the registration procedure 380 is configured to perform multi-modality image fusion to align images from different modalities acquired by the multi-domain data acquisition procedure 350 .

[0246] In one or more embodiments, the registration procedure 380 is configured to perform dynamic image sequence analysis to stack still images acquired at different time steps from a dynamic image sequence typically used to capture and quantify motion of anatomical structures such as the respiratory / cardiac systems.

[0247] In one or more embodiments, the registration procedure 380 is configured to perform data interpolation techniques to obtain biomarker values ​​at inter-cell center or inter-cell locations, which may include deterministic and / or statistical interpolation techniques.

[0248] Non-limiting examples of data interpolation techniques include nearest neighbor interpolation, linear interpolation, spline interpolation, polynomial interpolation, Lagrangian interpolation, Gaussian interpolation, Fourier transform, and wavelet transform.

[0249] In one or more embodiments, the registration procedure 380 receives additional data used during the multi-domain data acquisition procedure 350. As a non-limiting example, if a mesh has been generated and modified in pre-processing and post-processing, the modification information may be sent to the registration procedure 380 to make registration easier.

[0250] In one or more embodiments, the registration procedure 380 is configured to determine, for each domain representation of the plurality of domain representations, a respective center point or a respective center line of one or more anatomical structures of interest in the domain representation. It will be appreciated that the center point / center line may be determined using different methods, including manual methods (e.g., by receiving user input), automatic methods, or a combination thereof.

[0251] The registration procedure 380 is configured to determine a separate correspondence rule between the initial mesh and each of the multiple domain representations. In one or more embodiments, the correspondence rule may be determined based on the center points and / or center lines of each domain representation.

[0252] Correspondence rules are determined for regions in the domain representation that correspond to regions (i.e., sets of nodes) in the initial mesh. As a non-limiting example, the domain representation may be an axial ultrasound image, and the registration procedure 380 may determine axial view and node coordinates in the mesh that correspond to regions in the axial ultrasound image. The registration procedure 380 then determines correspondence rules between pixels in the axial ultrasound image and nodes in the mesh.

[0253] The individual correspondence rules may include a function mapping element from a given domain representation to a corresponding element (set of nodes) in the initial 3D mesh, allowing biomarker data associated with elements of the given domain representation to be mapped as features on the initial mesh. In other words, the individual correspondence rules are functions that describe the mapping from an individual coordinate system of the domain representation to the coordinate system of the initial mesh.

[0254] Since the initial 3D mesh and each domain representation may be based on different types of meshes and therefore may have different biomarker data densities for a given anatomical substructure, it turns out that it is necessary to determine the correspondence between structural elements, i.e., the location and number of elements in the individual domain representations that correspond to a given node at a given location in the initial 3D mesh.

[0255] The registration procedure 380 calculates correspondence rules between elements in the initial mesh and elements in a set of segmented images that include multiple anatomical segments. In one or more embodiments, the correspondence rules may be initially determined such that the set of segmented images is used as a "baseline" visual representation (i.e., baseline mask) of one or more anatomical structures of the patient that are encoded into the 3D parametric mesh.

[0256] The alignment procedure 380 calculates correspondence rules between elements in the initial mesh and elements in the structural mechanics representation. As a non-limiting example, the alignment procedure 380 may determine regions in the initial mesh of 10,000 nodes that correspond to regions in a surface mesh of 4,000 triangular shell elements of the structural mechanics representation and determine correspondence rules between data associated with the triangular elements and the nodes of the mesh. As a non-limiting example, the regions may include nodes located on different concentric 3D mesh layers.

[0257] The alignment procedure 380 calculates correspondence rules between elements in the initial mesh and elements in the fluid dynamics representation. As a non-limiting example, the alignment procedure 380 can determine which regions in the initial mesh of 10,000 nodes in the fluid dynamics representation correspond to which regions in a volumetric mesh of 4,000,000 tetrahedral elements, and determine correspondence rules between data associated with the tetrahedral elements and the nodes of the mesh. As a non-limiting example, the regions may include nodes located on different concentric 3D mesh layers.

[0258] The registration procedure 380 calculates correspondence rules between elements in the initial mesh and elements in the descriptive variable representation. As a non-limiting example, the registration procedure 380 can determine which regions in the initial mesh of 10,000 nodes correspond to which regions of pixels in the ultrasound image and determine correspondence rules between data associated with the pixels and nodes of the mesh. As a non-limiting example, the regions may include nodes located on different concentric 3D mesh layers.

[0259] As a non-limiting example, the alignment procedure 380 may be implemented by determining whether the nodes in the initial mesh are aligned with multiple elements p e After determining the anchor point, the registration procedure 380 may determine that, for biomarker data, the feature (i.e., biomarker) value at that node corresponds to p e Correspondence rules between multiple elements in a given domain representation can be generated by specifying that a weighted average of the biomarkers of

[0260] Additionally or alternatively, for elements in a given domain representation that do not directly correspond to nodes on an initial 3D mesh comprising multiple concentric 3D mesh layers, the registration procedure 380 may use distance to weight biomarker values ​​in the correspondence rule.

[0261] The alignment procedure 380 outputs a separate correspondence rule for each domain representation.

[0262] Parametric Mesh Encoding Procedure The parametric mesh encoding procedure 390 is configured to, among other things, (i) receive an initial 3D mesh; (ii) receive a multi-domain representation from the multi-domain data acquisition procedure 350; (iii) receive correspondence rules from the alignment procedure 380; (iv) use the correspondence rules to determine, for at least one given region, individual feature sets from individual biomarkers in the multi-domain representation; and (v) assign the feature sets to the at least one given region of the initial 3D mesh to obtain a 3D parametric mesh, wherein each node of the at least one region of the parametric mesh is associated with an individual plurality of feature channels that include at least the feature set.

[0263] The goal of the parametric mesh encoding procedure 390 is to use the initial 3D mesh to generate a 3D parametric mesh that is a single representation of the patient's body anatomy, including all biomarker data from the multiple domain representations. It will be appreciated that the biomarker data may be encoded in time onto the initial mesh. The 3D parametric mesh may then be provided for display on a user interface and used to render, display, and interact with visual representations of the patient-specific data, including displaying different views in 2D, 3D, and 4D, and displaying the multi-domain biomarker data encoded within the nodes of the 3D parametric mesh.

[0264] Determining characteristics The parametric mesh encoding procedure 390 is configured to determine a distinct set of feature values ​​from the given domain representation for each region of the given domain representation that has a corresponding region in the mesh using a distinct correspondence rule.

[0265] The set of distinct features corresponds to at least a portion of the biomarkers in a given domain representation. It will be appreciated that the set of distinct features extracted may vary depending on the region, concentric mesh layer, and domain representation. The features may include all biomarkers from a given domain representation, or may include only biomarkers of interest from a given domain representation (which may be predetermined by an operator).

[0266] The parametric mesh encoding procedure 390 uses individual correspondence rules to convert biomarker values ​​into nodal feature values ​​that are assigned to each node in the corresponding region in the initial 3D mesh.

[0267] The parametric mesh encoding procedure 390 is configured to encode structural information from a set of segmented images including multiple anatomical segments as features within a mesh. Structural information (i.e., locations) of segments of a given patient's anatomy are encoded into nodes of an initial 3D mesh, resulting in a parametric 3D mesh that can be used to render 2D and / or 3D representations of the physical structure of the corresponding anatomy displayed in the given patient's image set. The parametric mesh encoding procedure 390 may encode structural information, such as the location of the anatomical structure's centerline, boundaries, and locations of substructures (e.g., locations of the lumen, outer wall, thrombus, and calcification in a segmented aorta), into the parametric 3D mesh. The structural information encoded as features within the nodes of the 3D parametric mesh is used to render the 3D parametric mesh on a user interface. It will be appreciated that structural information can be encoded in time (if available), so that changes in the parametric 3D mesh over time can also be visually represented (e.g., the location of the lumen, outer wall, thrombus, and calcification at various times during the cardiac cycle).

[0268] The parametric mesh encoding procedure 390 is configured to encode the biomarkers of corresponding elements in the structural mechanics representation to associated nodes of the mesh using distinct correspondence rules. Each structural mechanics representation biomarker is encoded as a separate feature in multiple feature channels of the corresponding node. It will be appreciated that feature positions within the multiple feature channels may be reserved for the structural mechanics biomarkers, e.g., the structural mechanics biomarkers may be encoded in channels 10-20 of each node.

[0269] The parametric mesh encoding procedure 390 is configured to encode biomarkers of corresponding elements of the fluid dynamics representation into associated nodes of the mesh using distinct correspondence rules. Each of the fluid dynamics representation biomarkers is encoded as a separate feature in multiple feature channels of the corresponding node. It will be appreciated that feature positions within the multiple feature channels may be reserved for the fluid dynamics biomarkers, e.g., structural dynamics biomarkers may be encoded in channels 21-30 of each node.

[0270] The parametric mesh encoding procedure 390 is configured to encode the biomarkers of corresponding elements in the descriptive variable representation into associated nodes of the mesh using distinct correspondence rules. Each descriptive variable representation biomarker is encoded as a separate feature in multiple feature channels of the corresponding node. It will be appreciated that feature positions within the multiple feature channels may be reserved for structural dynamics biomarkers, e.g., structural dynamics biomarkers may be encoded in channels 31-50 of each node.

[0271] It will be appreciated that all biomarker data of interest for a given region of interest in the anatomy can thus be readily stored and retrieved using a single coordinate system of the parametric 3D mesh. Thus, for a given node corresponding to a region in the anatomy, e.g., a peripheral point on the ascending aorta, biomarker data from all modalities and physical representations associated with that given node can be encoded as features such as, for example, location, time, pixel intensity, strain values ​​including maximum principal strain, minimum principal strain, circumferential strain, and longitudinal strain, deformation values, fluid dynamics data, etc.

[0272] The parametric mesh encoding procedure 390 outputs a 3D parametric mesh composed of multiple concentric 3D mesh layers, each having a predetermined number of nodes, each containing multiple feature channels. The 3D parametric mesh is a patient-specific representation of the patient's anatomy, allowing for intuitive and systematic reporting of information from multiple domains on anatomically relevant maps extracted from the patient's original vascular scan.

[0273] Although the registration procedure 380 and the parametric mesh encoding procedure 390 have been described as separate procedures, it will be understood that such description is for illustrative purposes only, and that the registration procedure 380 and the parametric mesh encoding procedure 390 may be combined.

[0274] In one or more embodiments, the parametric mesh may be stored in a non-transitory storage medium or database 235 of the server 230 .

[0275] In one or more embodiments, the parametric mesh may be output and transmitted.

[0276] As a non-limiting example, the parametric mesh may be transmitted to a workstation computer 215 for display.

[0277] The parametric mesh may be displayed using any suitable 2D or 3D rendering technique known in the art and may be interacted with to visualize data from multiple feature channels.

[0278] The parametric mesh can thus provide a database containing all structural, functional, and descriptive data for the anatomical structures of interest for a given patient, so that the data encoded in the form of features at each node location can be rapidly retrieved and displayed for analysis.

[0279] The parametric mesh generation procedure 300 is repeated for multiple patients to obtain separate parametric meshes that each encode separate biomarkers for a separate patient.

[0280] The set of parametric meshes may be used to train various machine learning ML models, and a given parametric mesh from the set of parametric meshes may be associated with an individual patient.

[0281] It is understood that because all ML models rely on the same data types, weights can be shared or retrained very minimally when new information domains are introduced. Modular modeling makes it possible to retrain new architectures or for new tasks by utilizing fewer weights (parameters) and requiring less retraining of these weights. Consequently, this makes it easier to obtain generalizable models starting with a smaller number of vascular scans.

[0282] FIG. 4 shows a non-limiting example of a perspective view of a rendering of a first parametric mesh 400 taken from the anterior left side of the aorta including the iliac arteries, in accordance with one or more non-limiting embodiments of the present technique.

[0283] FIG. 5A shows a perspective view of a rendering of the first parametric mesh 400 of FIG. 4 with the top portion removed according to line 11, revealing multiple concentric 3D mesh layers 440 beneath the first parametric mesh 400.

[0284] FIG. 5B shows an enlarged detailed view of the multiple concentric 3D mesh layers 440 of FIG. 5A, including a selected node 454 and its multiple feature channels 460.

[0285] In this illustrated example, the plurality of concentric 3D mesh layers 440 includes ten layers (not all of which are numbered): a first concentric 3D mesh layer 442, a second concentric 3D mesh layer 444, a third concentric 3D mesh layer 446, ..., a ninth concentric 3D mesh layer 448, and a tenth concentric 3D mesh layer 450.

[0286] The first concentric 3D mesh layer 442 is the innermost layer and represents the core (not numbered) of the first parametric mesh 400 .

[0287] The tenth concentric 3D mesh layer 450 is the outermost layer and represents the outer wall of the aorta.

[0288] As a non-limiting example, the first concentric 3D mesh layer 442 can be used to encode features such as the diameter and curvature of the aortic centerline, while the tenth layer 450 can be used to encode wall distortion, shape, and the presence of calcification. The intermediate concentric layers, i.e., the second concentric 3D mesh layer 444, the third concentric 3D mesh layer 446, ..., the ninth concentric 3D mesh layer 448, can be used to encode the presence of thrombus, flow patterns, image pixel color, etc., derived from biomarkers in multiple domains (e.g., different imaging modalities, computational fluid dynamics, etc.).

[0289] Although the tenth concentric 3D mesh layer 450 is shown as the outermost layer of the first parametric mesh 400, it will be understood that in one or more other embodiments, the first parametric mesh 400 may include one or more additional layers located outside the represented anatomical structure (e.g., outside the outer wall of the aorta). Such additional layers may or may not correspond to other anatomical structures and may be used to encode additional information.

[0290] FIG. 6 shows a top or axial plan view of the first parametric mesh 400 of FIG. 5B rendered along line 11 with the top portion removed.

[0291] 5B and 6, individual nodes 454 of the ninth concentric 3D mesh layer 448 of the first parametric mesh 400 are associated with respective node coordinates 456 represented as (m,n,p), where m is the circumferential coordinate, n is the longitudinal coordinate, and p is the time frame coordinate. Each node 454 of the mesh 400 has multiple feature channels 460 encoding features derived from different biomarkers. The multiple feature channels 460 may include, for example, structural features 472, hydrodynamic features 476, and variable descriptive features 478.

[0292] FIG. 7 shows a perspective view of a rendering of a second parametric mesh 700 taken from the anterior left side of the aorta and iliac arteries, in accordance with one or more non-limiting embodiments of the present technology.

[0293] The second parametric mesh 700 includes an aortic mesh 720 , a left iliac artery mesh 724 , and a right iliac artery mesh 726 .

[0294] In the illustrated second parametric mesh 700, only the outermost or surface concentric 3D mesh layers are shown, with each axial layer represented by an elliptical shape for each discrete axial value (i.e., each longitudinal value). A centerline 730 (corresponding to a node in the first 3D layer of the aortic mesh 720) extends within the aortic mesh 720 and bifurcates into a left iliac centerline 734 (corresponding to a node in the first 3D layer of the left iliac mesh 724) extending within the left iliac mesh 724, and a right iliac centerline 736 (corresponding to a node in the first 3D mesh layer of the right iliac mesh 726) within the right iliac mesh 726.

[0295] A second aortic line 740 extends within the aortic mesh 720 and is defined by nodes having the same circumferential coordinate m but different longitudinal coordinates n (i.e., located on different axial layers). The second aortic line 740 bifurcates into a second left iliac line 744 and a second right iliac line 746. Like the nodes in the second aortic line 740, the nodes in each of the second left iliac line 744 and the second right iliac line 746 have the same circumferential coordinate m in their respective arrangements but different longitudinal coordinates n in their respective arrangements (i.e., located on different axial layers).

[0296] FIG. 8 shows a schematic diagram of a first user interface 800 displaying a third parametric mesh 805, a corresponding third parametric mesh array 850, and visual renderings of user interface elements in the form of a layer slider 870, a domain slider 880, and a substructure slider 890.

[0297] A third parametric mesh 805 is 3D rendered on the left side of the user interface 800. The third parametric mesh 805 includes an aortic mesh 810, a left iliac artery mesh 830, and a right iliac artery mesh 840. It will be appreciated that the visual representation of the third parametric mesh 805 is generated from the structural features encoded in the nodes of the third parametric mesh 805, and visually represents the particular anatomy of a given patient from which multi-domain information was extracted.

[0298] A third parametric mesh array 850 is displayed in the upper right portion of the user interface 800. The third parametric mesh array 850 includes an aortic mesh array 852, a left iliac artery array 856, and a right iliac artery array 858.

[0299] The third parametric mesh array 850 represents the third parametric mesh 805 unwrapped relative to the centerline (or vertical axis), with the columns of the outer layer array 850 corresponding to the vertical node positions (i.e., positions along the vertical axis) on the third parametric mesh 805 and the rows of the third parametric mesh array 850 corresponding to the circumferential node positions (i.e., positions along the circumference). Each array element in the third parametric mesh array 850 corresponds to a distinct node on a visual rendering of the third parametric mesh 805.

[0300] In the illustrated non-limiting example, it can be seen that necks 820 in aortic mesh 810 are represented by neck array 854 in aortic mesh array 852, left iliac artery mesh 830 is represented by left iliac artery array 856, and right iliac artery mesh 840 is represented by right iliac artery array 858.

[0301] Each node in the outer layer of the third parametric mesh 805 can be accessed in the third parametric mesh array 850 using the same coordinates (M,N) as above, where M is the circumferential node coordinate and N is the longitudinal node coordinate. It will be appreciated that each concentric mesh layer of the third parametric mesh 805 can be represented as a separate array of the same size as the third parametric mesh array 850, and each time frame (i.e., corresponding to the parametric mesh 805 at a different point in time) can be represented as a respective array of the same size.

[0302] The layer slider 870 allows for the selection and display of different concentric 3D mesh layers of the third parametric mesh 805, including the innermost core layer, wall layers, and outer layers, as well as layers located in between. By selecting a layer using the layer slider 870, structural information encoded in the nodes of that layer may be processed to render a graphical representation of the selected mesh layer.

[0303] The graphical rendering displayed on the left changes depending on the layer selected in the layer slider 870, but the size and structure of the corresponding third parametric mesh array 850 remains the same, i.e., it contains the same number of nodes or elements.

[0304] The domain slider 880 allows for the selection and display of different domains encoded in the third parametric mesh 805, including, by way of non-limiting examples, strain, computational fluid dynamics (CFD), and calcification. By selecting a domain using the domain slider 880, biomarkers encoded as features at the nodes may be processed to render a graphical representation of the selected domain. The user interface 800 also includes a substructure slider 890 in the form of a neck parameter slider that allows for the display of different features specific to the neck in the aorta.

[0305] Although not shown in FIG. 8, the user may also use other interface elements to select different types of rendering and projection.

[0306] FIG. 9A shows a schematic diagram of the second user interface 900 showing a visual representation of a fourth parametric mesh 905 of the aorta including the iliac arteries, where the concentric mesh layer 915 between the lumen and wall is highlighted after being selected with the layer slider 930, and the corresponding layer array 920 shows grayscale pixel intensities after the "grayscale" domain is selected with the domain slider 940.

[0307] FIG. 9B shows a schematic diagram of the second user interface of FIG. 9A, in which the core concentric mesh layer 965 selected with the layer slider 930 is highlighted, and the corresponding layer array 970 displays the grayscale pixel intensities of the core concentric mesh layer 965 with the “grayscale” domain selected with the domain slider 940.

[0308] Having described the parametric mesh generation procedure 300 with reference to Figure 3 and various examples of parametric meshes with reference to Figures 4-9B, reference is now made to Figures 10A and 10B, which show a flowchart of a method 1000 for generating a parametric mesh in accordance with one or more non-limiting embodiments of the present technology.

[0309] It will be appreciated that the procedure(s) of the parametric mesh generation procedure 300 may be integrated into the method 1000 .

[0310] Instructions In one or more embodiments, server 230 comprises at least one processor, such as processor 110 and / or GPU 111, operably connected to a non-transitory computer-readable storage medium, such as solid-state drive 120 and / or random access memory 130, that stores computer-readable instructions. The at least one processor is configured to or operable to perform method 1000 upon execution of the computer-readable instructions.

[0311] The method 1000 begins at process step 1002 .

[0312] According to processing step 1002, at least one processor receives an image set of a given patient acquired by a medical imaging device, the image set including at least one image of at least some of the anatomical structures within the given patient's body.

[0313] In one or more embodiments, the image set is acquired by a medical imaging device 210 .

[0314] In one or more other embodiments, the image set may include multiple images in the form of an image stack. In one or more alternative embodiments, the image set includes multiple images in the form of a multi-phase stack.

[0315] In one or more embodiments, the anatomical structure includes the aorta of a given patient. Additionally, the anatomical structure may include the iliac arteries of a given patient.

[0316] According to process step 1004, at least one processor segments the image set to obtain a plurality of anatomical segments of at least some of the internal anatomy of a given patient.

[0317] In one or more embodiments, at least one processor obtains a plurality of anatomical segments or segmented tissues using manual and / or automatic segmentation methods.

[0318] In one or more embodiments, the processor uses a set of trained segmentation models 260 to segment a set of images to obtain a plurality of segments.

[0319] As a non-limiting example, the set of trained segmentation models 260 may be trained to segment an aortic region including one or more of the aorta and iliac arteries, where the segmented aortic region includes an ROI including the lumen, the aortic wall, the ILT (if present), and calcification (if present).

[0320] In one or more embodiments, processing steps 1002 and 1004 may be replaced by a single processing step in which a processor receives multiple anatomical segments from another processor.

[0321] According to processing step 1006, at least one processor receives an initial 3D mesh for representing an anatomical structure, the initial 3D mesh including a plurality of concentric 3D mesh layers, each of the plurality of concentric 3D mesh layers including the same predetermined number of nodes.

[0322] In one or more embodiments, each node has at least one distinct time coordinate that allows the node and the mesh to be represented in time.

[0323] The processor initializes the mesh according to a set of mesh parameters that specify at least the geometry and number of nodes of the mesh, where each predetermined region in the mesh may be, for example, part of one or more anatomical structures of interest and may correspond to at least a portion of a segmented anatomical segment obtained by segmenting the image set in processing step 1004.

[0324] In one or more embodiments, process step 1006 may be performed before process steps 1002 and 1004 .

[0325] According to processing step 1008, at least one processor determines at least one region in the mesh that corresponds to a given anatomical segment of the plurality of anatomical segments to obtain a correspondence rule between the at least one region in the mesh and the given anatomical segment.

[0326] In one or more embodiments, the correspondence rule represents a functional mapping element from a region in a given anatomical segment of the plurality of anatomical segments to a corresponding region of a node in the initial mesh.

[0327] According to processing step 1010, at least one processor uses a correspondence rule to encode at least one node set of the 3D mesh with a distinct set of features from at least one distinct anatomical segment to obtain a 3D parametric mesh, wherein each node of the at least one node set in the 3D parametric mesh is associated with a distinct plurality of feature channels comprising the distinct set of features.

[0328] In one or more embodiments, processing step 1010 includes determining a set of distinct features from biomarkers in at least one distinct anatomical segment using distinct correspondence rules, and assigning a distinct set of features from the at least one distinct anatomical segment to each of the at least one distinct set of nodes.

[0329] In one or more embodiments, at least one processor uses correspondence rules to extract biomarkers, such as locations of anatomical segments and pixel intensities from the anatomical segments.

[0330] It will be appreciated that because the discretized geometry of the initial mesh may differ from the discretized geometry of the multiple segments in the image, the correspondence rule may specify which regions and cell locations within the multiple segments correspond to nodes in the mesh and how the cell values ​​should be reported to the initial 3D mesh. As a non-limiting example, each node may correspond to an area of ​​four pixels in the segmented image, and therefore the correspondence rule may specify that the biomarker values ​​of the four pixels should be averaged to obtain a feature. Thus, the processor determines features based on the biomarker values ​​of each node based on the correspondence rule and populates each node with the determined features.

[0331] The processor encodes or populates a set of features of at least one region within the segment into nodes of the mesh. The set of features may include, for example, location and pixel intensity. Thus, the 3D parametric mesh provides a discretized 2D or 3D representation of the regions within the anatomical segment that are encoded as nodal features. In one or more embodiments, temporal information may be encoded as features at the nodes of the parametric mesh, allowing for visualization of the evolution of the features over time. The nodal features may be used to generate a visual representation of the 3D parametric mesh and may be displayed in one or more arrays corresponding to the 3D parametric mesh.

[0332] It will be appreciated that processing steps 1008-1010 may be repeated for other regions and segments such that all necessary information about the anatomical structure of interest is encoded in the 3D parametric mesh.

[0333] According to processing step 1012, at least one processor receives a domain representation including biomarkers associated with internal anatomical structures of a given patient.

[0334] In one or more embodiments, at least one processor receives at least one of a structural mechanics representation including structural mechanics biomarkers, a fluid mechanics representation including fluid mechanics biomarkers, and a descriptive variable representation including variable descriptive biomarkers.

[0335] Each domain representation may have a different data format and / or data density and / or resolution.

[0336] In one or more embodiments, the given domain representation may be a mesh, such as a polygonal mesh. A polygonal mesh includes vertices, edges, and faces. The faces may include one of triangles (triangle mesh), quadrilaterals (quads), convex polygons (n-gons), concave polygons, and polygons with holes. The mesh may be a 2D or 3D mesh, with or without a temporal component.

[0337] In one or more other embodiments, the given domain representation may be a 2D or 3D image.

[0338] In one or more embodiments, the processor uses a registration technique to determine another correspondence rule.

[0339] The individual correspondence rules may include a functional mapping element from a given domain representation to a corresponding element in the initial mesh, thereby allowing biomarker data associated with elements of the given domain representation to be mapped as features on the initial mesh. In other words, the individual correspondence rules are functions that describe the mapping from the individual coordinate system of the domain representation to the coordinate system of the initial mesh.

[0340] According to processing step 1014, the at least one processor determines another region in the domain representation and another region in the parametric mesh that corresponds to another given anatomical segment to obtain another correspondence rule.

[0341] It will be appreciated that the other region may be the same region as in process step 1008 or a different region.

[0342] According to processing step 1016, the at least one processor uses another individual correspondence rule, at least one other individual set of nodes in the 3D parametric mesh with another feature set based on the individual biomarkers, and each node of the at least one other individual set of nodes in the 3D parametric mesh is associated with multiple individual feature channels comprising the other feature set.

[0343] In one or more embodiments, the number of features or features represented in the feature channels of each node in the 3D parametric mesh may vary.

[0344] In one or more embodiments, to perform processing step 1016, at least one processor determines a different set of features from biomarkers associated with a given anatomical segment in the domain representation based on a different correspondence rule and assigns the different set of features to each node of at least one other distinct set of nodes in the 3D parametric mesh.

[0345] It will be appreciated that processing steps 1014-1016 may be performed multiple times, each for a different domain representation containing a distinct biomarker.

[0346] The method 1000 then ends.

[0347] AAA growth training data generation procedure

[0348] Referring to FIG. 11, a AAA growth training data generation procedure 1100 is shown in accordance with one or more non-limiting embodiments of the present technology.

[0349] The purpose of the AAA growth training data generation procedure 1100 is to generate labeled training data indicative of a patient's AAA growth based on a parametric mesh mask. The parametric mesh mask corresponds to a parametric mesh generated for the same patient at subsequent time periods (e.g., the patient's follow-up imaging sessions at 1 month, 6 months, 1 year, etc.) and encoded with biomarker data for the subsequent time periods. The data can then be labeled and provided as training for a set of growth prediction models 270 for predicting AAA growth.

[0350] The use of masked parametric meshes allows for the acquisition and comparison of feature data according to sections and / or arbitrary geometric reference points (e.g., the centerline of the aorta) on 2D and / or 3D parametric meshes.

[0351] It will be appreciated that in one or more alternative embodiments of the present technology, the AAA growth training data generation procedure 1100 can be adapted and used to predict the growth of other types of aneurysms in blood vessels, such as thoracic aneurysms.

[0352] The AAA growing training data generation procedure 1100 includes, among other steps, a parametric mesh generation procedure 1110, a mesh feature comparison procedure 1160, a training data generation procedure 1180, and a model training procedure 1190.

[0353] In one or more embodiments, a parametric mesh generation procedure 1110 is performed to obtain a parametric mesh encoded with features of the patient's fluid dynamics biomarkers, structural dynamics biomarkers, and descriptive biomarkers. In one or more embodiments, the fluid dynamics biomarkers may include wall shear stress (i.e., TAWSS), and the structural dynamics biomarkers may include geometric biomarkers indicative of the geometry of the ROI and ILT thickness. The ROI may include the thoracic region (e.g., ascending aorta, aortic arch, descending thoracic aorta) and / or abdominal aortic region (e.g., suprarenal abdominal aorta, infrarenal aorta, renal arteries, lumbar arteries) and iliac arteries (e.g., common iliac arteries, external iliac arteries, internal iliac arteries).

[0354] The structural mechanics biomarkers encoded into the mesh may also include feature biomarkers indicative of strain (e.g., maximum principal strain, minimum principal strain, circumferential strain, and longitudinal strain, corresponding strain rate, peak strain, and RAW).

[0355] The parametric mesh generation procedure 1110 may be similar to the parametric mesh generation procedure 300 .

[0356] In this embodiment, the parametric mesh generation procedure 1110 is configured to perform a mask encoding procedure 1150 .

[0357] Parametric mesh generation procedure The mask encoding procedure 1150 is configured to receive, for a given patient, one or more follow-up images of the patient's body, the one or more follow-up images being acquired during subsequent imaging sessions after a baseline imaging session. The patient may, for example, be a patient with AAA.

[0358] The parametric mesh generation procedure 1110 is configured to perform an implementation of the parametric mesh generation procedure 300 based on one or more follow-up images of the patient, where biomarkers generated based on the follow-up image sessions are encoded as features. In such an implementation, the features generated for the follow-up imaging sessions may be encoded into the parametric mesh, including information obtained regarding the geometric evolution of the vessels in the form of relative mask differences between the baseline mask and the follow-up mask that are recorded in the parametric mesh.

[0359] The parametric mesh follow-up mask represents a parametric mesh encoded with feature channels generated from biomarkers based on the patient's follow-up imaging sessions. In such an embodiment, the parametric mesh generation procedure 1110 performs a mask encoding procedure 1150. The mask encoding procedure 1150 can include a segmentation procedure similar to the segmentation procedure 340, a multi-domain data acquisition procedure similar to the multi-domain data acquisition procedure 350, a registration procedure similar to the registration procedure 380, and a parametric mesh encoding procedure similar to the parametric mesh encoding procedure 390.

[0360] In some embodiments, the features encoded in the follow-up mask of the parametric mesh may be a subset of the features encoded in the baseline mask of the parametric mesh, i.e., not all of the same features (e.g., fluid dynamics, structural dynamics biomarkers, and descriptive variable representation biomarkers) may be encoded based on the follow-up imaging session.

[0361] The parametric mesh generation procedure 1110 outputs a parametric mesh for the patient that includes a mask encoding the biomarkers for the baseline and follow-up imaging sessions.

[0362] The parametric mesh is stored in a storage medium such as a database 235 or the memory of the computing device 100 .

[0363] It will be understood that the parametric mesh with masks at baseline and follow-up sessions encodes patient information between pixels, surfaces, and patient space. Therefore, variations in the encoded features (i.e., biomarkers) between the baseline and follow-up states of the patient's body in the parametric mesh can be obtained for visualization, analysis, and comparison. It will be understood that changes in geometric and functional biomarkers (i.e., fluid dynamics and structural dynamics biomarkers) can indicate AAA growth.

[0364] In one or more embodiments, the parametric mesh generation procedure 1110 is configured to generate different visual representations based on masks of the parametric meshes to indicate differences in selected features between the masks of the parametric meshes. In one or more embodiments, the parametric mesh generation procedure 1110 generates views of the parametric mesh based on transverse, sagittal, and parasagittal planes, as well as oblique planar views. The different visual representations may be provided for display on a graphical user interface of a display of a computing device for interaction with a user (e.g., a medical professional).

[0365] As a non-limiting example, the parametric mesh generation procedure 1110 may generate one or more of an anterior view, a posterior view, a lateral view, a cross-sectional view, and a longitudinal cross-sectional view of the segmented portion of the aorta based on the mask of the parametric mesh. Different views may be generated based on the data array.

[0366] The AAA growth training data generation procedure 1100 is configured to perform a mesh feature comparison procedure 1160 .

[0367] Mesh feature comparison procedure A mesh feature comparison procedure 1160 is configured to compare selected features on selected nodes between masks of the patient's parametric mesh.

[0368] In one or more embodiments, an indication of one or more of the selected features, selected regions in 2D and 3D (i.e., selected node sets) for the mesh feature comparison procedure 1160 may be provided by a user via an input / output interface of a computing device. In one or more other embodiments, an indication of one or more of the selected features, selected regions in 2D and 3D (i.e., selected node sets) for the mesh feature comparison procedure 1160 may be received from a non-transitory storage medium and / or may be received from database 235.

[0369] In some implementations, the mesh feature comparison procedure 1160 performs a comparison of selected domain features between masks of parametric meshes, which may be represented as arrays.

[0370] It will be appreciated that comparison of features encoded in the parametric meshes can be easily performed because the biomarkers extracted or generated based on the patient's baseline and follow-up images have previously been registered and encoded into a mask on the same parametric mesh.

[0371] In one or more embodiments, the mesh feature comparison procedure 1160 is configured to compare baseline and follow-up mask features encoded in one or more selected regions of the plurality of 3D concentric mesh layers and selected anatomical structures.

[0372] A mesh feature comparison procedure 1160 can perform a comparison of features on the mask between different planes and structures of the selected anatomical segment of the parametric mesh.

[0373] In some alternative implementations, the selected nodes and features on the parametric mesh may correspond to all nodes and all features encoded on the parametric mesh.

[0374] As a non-limiting example, the mesh feature comparison procedure 1160 can be configured to compare the baseline mask features and follow-up mask features within a 3D mesh layer against an outer mesh representing the outer surface of the anatomical structure, and / or against an innermost mesh at the centerline of the anatomical structure, and / or against further layers further outside the anatomical structure.

[0375] As a non-limiting example, the anatomical structure selected for the mask comparison procedure may be defined as the node corresponding to the abdominal aorta from below the celiac trunk to the common iliac bifurcation, used as a landmark to ensure the same portion of the artery is assessed at baseline and follow-up, and the selected segmented region of interest may include the patient's aortic wall and lumen. The selected anatomical structure and segmented region are referenced using appropriate sequence, nodal, and time coordinates.

[0376] In one or more embodiments, if available, the mesh feature comparison procedure 1160 can perform comparison measurements based on one or more of the aortic length, cross-sectional area, tortuosity, and volume measurements of the regions of the parametric mesh.

[0377] The selected features are biomarkers encoded as features on the parametric mesh and are used for comparison between masks of the parametric mesh.

[0378] The selected biomarkers may include function-based or biomechanics-based biomarkers that are encoded as features of the nodes of the parametric mesh. The selected biomarkers may have been received or generated during the parametric mesh generation procedure 1100. In one or more other embodiments, the selected features may be generated using biomarkers encoded in the parametric mesh.

[0379] The selected mesh features or biomarkers may include one or more of fluid dynamics biomarkers, structural dynamics biomarkers, and descriptive variable biomarkers.

[0380] In some embodiments, the selected features include geometric features between the parametric mesh-encoded masks, including the shape and size of the segmented tissue (e.g., diameter, length, curvature (relative to a centerline or reference node)), as well as the location, size, and angle of branches (e.g., carotid, subclavian, and renal arteries). Additionally, geometric features may include irregularities, radii of curvature, tortuosity, and asymmetry of the aneurysm wall of the anatomical structure or region.

[0381] In one or more embodiments, the geometric characteristics may include 2D and / or 3D distances relative to the centerline of the parametric mesh or relative to another reference point (ie, a node).

[0382] In some implementations, the mesh feature comparison procedure 1160 is configured to compare shape and texture features between masks encoded with parametric meshes.

[0383] The selected features may include one or more of time-averaged wall shear stress (TAWSS), in vivo principal strain, and ILT thickness. In one specific, non-limiting example, the selected biomarkers include time-averaged wall shear stress (TAWSS), in vivo principal strain, and ILT thickness.

[0384] Additionally, the selected features may include patient information such as age, biological sex, weight, height, family history of AAA, smoking history, cardiac disease, hypertension (HTN), chronic obstructive pulmonary disease (COPD), and diabetes mellitus (DM).

[0385] A mesh feature comparison procedure 1160 is configured to compare selected features in 3D between the baseline and follow-up masks of the parametric mesh. Comparing the selected features in 3D allows for evaluation of changes in the patient's anatomy and may correspond to features indicative of AAA growth.

[0386] In one or more embodiments, the mesh feature comparison procedure 1160 is performed on biomarkers encoded as features at a particular instant in time within the parametric mesh. As a non-limiting example, the mesh feature comparison procedure 1160 may be performed on the diastolic phase of the cardiac cycle, where biomarkers on a mask of the parametric mesh are compared for the diastolic phase of the cardiac cycle. It will be appreciated that such an embodiment may enable training of an ML model that predicts AAA growth from static rather than dynamic medical images, minimizing inconvenience to patients (such as radiation dose in a CT scan), facilitating the image acquisition process, and conserving computational resources (i.e., minimizing processing time and saving storage space).

[0387] As a non-limiting example, the nodes may correspond to nodes encoding values ​​of intraluminal thrombus (i.e., the difference measured using the aortic wall surface mesh and the luminal surface mesh) and the patient's TAWSS on the parametric mesh, and the mesh feature comparison procedure 1160 may perform a comparison of the thrombus thickness and TAWSS between masks of the parametric mesh.

[0388] In some implementations, the mesh feature comparison procedure 1160 determines, for each selected feature and selected region in the parametric mesh, an individual comparison value based on a comparison between masks of corresponding nodes of the parametric mesh. As a non-limiting example, the individual comparison value corresponds to the difference between each selected biomarker for the mask. Optionally, the difference between features can be calculated temporally (i.e., over at least a portion of a cardiac cycle) by comparing corresponding time steps encoded in the masks. Alternatively, for features expressed in time, the difference may be calculated based on calculated time average values.

[0389] In one or more other embodiments, the mesh feature comparison procedure 1160 is configured to perform a comparison of the selected features and selected regions between masks of the parametric mesh and assign normalized values ​​as individual comparison values.

[0390] In one or more embodiments, the mesh feature comparison procedure 1160 is configured to perform a comparison of selected features and regions between masks of the parametric meshes and assign a binary or continuous value as the respective comparison value. The binary value may be determined, for example, by comparing the difference in selected feature values ​​to a threshold value; in response to a difference equal to or greater than the threshold, the mesh feature comparison procedure 1160 may assign one binary value (e.g., 1), and in response to a difference less than the threshold, the mesh feature comparison procedure 1160 may assign another binary or continuous value (e.g., 0). As a non-limiting example, the mesh feature comparison procedure 1160 may compare nodal biomarkers on axial and circumferential sections perpendicular to the aortic center and assign a binary value based on a threshold difference (e.g., 2.5 cm).

[0391] In one or more alternative embodiments, the mesh feature comparison procedure 1160 is configured to perform a comparison of selected features and regions between the masks of the parametric mesh and assign multi-class values ​​as individual comparison values, which may be assigned based on threshold ranges for each selected feature.

[0392] Referring briefly to FIG. 13 , a baseline imaging session 1302 is shown that is used to acquire a wall-tracking mesh 1304, a lumen computational model 1306, and an in vivo strain analysis 1308, in addition to a CFD simulation (not shown) and wall-to-lumen distance measurements (not shown). This biomarker data is generated and / or acquired during the multi-domain data acquisition procedure 350 and encoded as features in a baseline mask of a parametric mesh (not shown). The data is used to generate biomarkers such as strain, ILT, and shear stress, which are also encoded as features in the baseline mask of a parametric mesh (not shown). Biomarker data from a subsequent follow-up imaging session 1322 is registered and encoded as features on a follow-up mask of a parametric mesh 1340, and local diameter growth 1350 is determined. The parametric mesh and the mask of local diameter growth 1320 are used as training data inputs for the growth prediction training procedure 1100.

[0393] Figure 12 shows a baseline data structure 1210 having data values ​​on the nodes of a baseline mask of a parametric mesh and a follow-up data structure 1220 having data values ​​on the nodes of a follow-up mask of a parametric mesh 1220. Figure 12 also shows a data comparison value 1230 between the baseline mask and the follow-up mask of a parametric mesh, in accordance with a non-limiting embodiment of the present technology.

[0394] In the baseline data structure 1210, region 1212 corresponds to values ​​on the outer wall nodes, and region 1214 corresponds to values ​​for the lumen and ILT. In the follow-up data structure 1220, region 1222 corresponds to values ​​on the outer wall nodes, and region 1224 corresponds to values ​​for the lumen and ILT. Data comparison values ​​1230 indicate structural changes between the baseline mask and the follow-up mask of the patient's parametric mesh.

[0395] Returning to FIG. 11, in some implementations, the mesh feature comparison procedure 1160 is configured to calculate the mean of a patch or region (ie, a set of nodes) of the structure of interest.

[0396] As a non-limiting example, the surface mesh defining each aorta on the parametric mesh is divided into 96 patches (12 axial sections and 8 circumferential sections perpendicular to the centerline of the aorta), three selected biomarkers, or local weakening (RW) components (TAWSS, strain, and ILT) are encoded as local (patch) averages to obtain local characteristics, and a mesh feature comparison procedure 1160 determines local diameter growth, calculated as the diameter difference at the level of each axial section.

[0397] 14 shows non-limiting examples of node distributions and patch-based average distributions of selected features (i.e., encoded biomarkers) on a parametric mesh. The selected features include TAWSS, strain, and ILT. The lumen surface of the parametric mesh is shown along with the node distribution of TAWSS 1410 and the patch-based average of the node distribution of TAWSS on the parametric mesh 1420. The wall surface of the parametric mesh is shown along with the node distribution of strain 1430, the patch-based average distribution of strain 1440, the node distribution of ILT 1450, and the patch-based average distribution of ILT 1460.

[0398] 15 shows the regional growth difference between masks of the parametric mesh 1500. The regional growth difference is assessed as a measure of regional diameter change, which was determined by performing a mesh feature comparison procedure 1160 on the geometric features at baseline and follow-up imaging sessions by comparing diameters at multiple sections of the node perpendicular to the aortic centerline 1520.

[0399] Returning to FIG. 11, in one or more embodiments, the mesh feature comparison procedure 1160 outputs, for each selected region and biomarker, a separate comparison value that may be indicative of AAA growth.

[0400] In some implementations, the mesh feature comparison procedure 1160 outputs an individual comparison value for each individual feature (ie, biomarker) for each node of the selected region of the parametric mesh.

[0401] The mesh feature comparison data may be represented as a data structure of comparison values ​​(eg, a matrix or tensor) of selected mesh features.

[0402] In one or more alternative embodiments, the mesh feature comparison data may be graphically rendered and displayed in a graphical user interface (GUI) for analysis by a medical professional.

[0403] The mesh feature comparison procedure 1160 outputs mesh feature comparison data calculated between masks of parametric meshes for a given patient.

[0404] The AAA growth training data generation procedure 1100 is configured to perform a parametric mesh generation procedure 1110 and a mesh feature comparison procedure 1160 for multiple patients.

[0405] Training data generation procedure The AAA growth training data generation procedure 1100 performs a training data generation procedure 1180. The training data generation procedure 1180 generates one or more training data sets for training a set of growth predictive models 270.

[0406] The training data generation procedure 1180 includes a labeling procedure 1185 .

[0407] Labeling Procedure In one or more embodiments, the labeling procedure 1185 is configured to associate the mesh comparison features and / or parametric meshes for each patient obtained during the mesh feature comparison procedure 1160 with a growth label indication. The growth label indication refers to a categorical or numerical annotation associated with each AAA instance for a patient in the dataset, indicating the presence, rate, and / or extent of aneurysmal expansion. The label serves as a target variable for a supervised learning algorithm, enabling the model to learn patterns associated with AAA growth.

[0408] The growth labels may include binary growth labels. In one or more other embodiments, the growth labels may include multi-class growth labels. In one or more alternative embodiments, for example, when predicting growth values ​​using a regression model, the training data generation procedure 1180 may associate the generated growth values ​​based on values ​​of mesh comparison features representing growth.

[0409] It will be appreciated that the growth labels may be provided by a medical professional and extracted from the parametric mesh.

[0410] In one or more other embodiments, the training data generation procedure 1180 is configured to label each of the mesh comparison features with a distinct label, where the distinct label is one of non-significant AAA growth and significant AAA growth. The labeling procedure 1170 may then associate the distinct labels with the parametric mesh and / or mesh comparison features of a given patient.

[0411] The training data generation procedure 1180 is configured to generate a training data set based on mesh comparison features between a mask of the parametric mesh and / or representation of features on the parametric mesh with representation of growth labels associated with the patient.

[0412] In one or more embodiments, the training data generation procedure 1180 is configured to generate multiple training data sets, each training data set generated for one or more of a different type of feature, a different type of region on the parametric mesh, and a different type of prediction task based on features within the parametric mesh.

[0413] In one or more embodiments, the training data generation procedure 1180 is configured to perform a feature selection procedure (not shown) to select features to be used in the training data set. A non-limiting example of a feature selection procedure is Boruta feature selection, although it will be understood that other alternative feature selection techniques may also be used.

[0414] Model training procedure The purpose of the training procedure 1190 is to train the growth prediction model 270 to perform growth prediction by classifying the baseline images as either indicative of AAA growth or not indicative of AAA growth based on the comparison features and associated labels. In one or more alternative embodiments, the training procedure 1190 is configured to train the growth prediction model 270 to perform growth prediction by classifying the baseline images as either indicative of AAA growth or not indicative of AAA growth based on the comparison features and associated labels.

[0415] The model training procedure 1190 is configured to train each of the set of growth prediction models 270 to perform growth predictions based on a training dataset with associated labels.

[0416] The growth prediction model 270 is trained to predict functional and structural features of the follow-up mask of the parametric mesh based on features encoded in the baseline mask of the parametric mesh.

[0417] In one or more implementations, the model training procedure 1190 is configured to use ensemble learning techniques to train the set of growing prediction models 270. As a non-limiting example, the model training procedure 1190 may be configured to use an ensemble decision tree forest, such as Extra-Trees. Extra-Trees is an ensemble ML approach that trains multiple decision trees and aggregates the results from the group of decision trees to output a growing prediction.

[0418] During training, the set of growing predictive models 270 performs individual classification of each baseline parametric mesh based on the provided features. A loss function is then used to calculate a loss based on the predictions and the labels associated with the baseline parametric mesh, and the parameters of the set of growing predictive models 270 are updated based on the calculated loss. This procedure is repeated iteratively until convergence and / or a stopping criterion is reached. In one or more embodiments, the model training procedure 1190 may stop upon reaching one or more of a desired performance threshold (e.g., accuracy of the classification task with minimal overfitting), a computational budget, a maximum training period, a lack of performance improvement, a system failure, etc.

[0419] The model training procedure 1190 may perform a testing and validation procedure to output a set of trained predictive models 270.

[0420] As a non-limiting example, the model training procedure 1190 may use 10-fold cross-validation.

[0421] The model training procedure 1190 is configured to store the set of trained growth predictive models 270. The model training procedure 1190 stores model parameters for the set of trained growth predictive models 270. In some implementations, the model training procedure 1190 transmits the set of trained growth predictive models 270 to another computing device.

[0422] One or more embodiments of the present technology allow predictions of AAA growth to be obtained from 3D maps and mapped directly to a parametric mesh without intermediate steps.

[0423] One or more embodiments of the present technology enable growth prediction in a multi-modality manner. For example, one or more models can be trained to predict growth from structural mechanics-based features or biomarkers such as TAWSS and ILT, and one or more models can be trained to predict growth using structural mechanics-based features (e.g., strain) and descriptive geometric features (e.g., image-based features) encoded in a parametric mesh.

[0424] One or more embodiments of the present technology may enable growth to be predicted based on the presence of structures outside the aorta, such as the patient's intercostal arteries, vertebrae, etc., encoded in the parametric mesh.

[0425] inference During inference, the set of trained growth prediction models 270 is configured to, among other things, (i) receive a parametric mesh generated based on a baseline medical imaging session, (ii) access the set of trained growth prediction models 270, and (iii) determine an individual predictive value indicative of growth based on features of the parametric mesh.

[0426] It will be appreciated that the set of trained growth prediction models 270 may include one or more classification models trained to predict growth in a binary manner (e.g., indicates growth or does not indicate growth) or a multi-class manner. Additionally or alternatively, the set of trained growth prediction models 270 may include one or more regression models trained to predict one or more values ​​related to growth (e.g., aneurysm expansion rate or extent of growth).

[0427] In one or more embodiments, the set of trained growth prediction models 270 is configured to perform growth prediction based on geometric features, ILT thickness, and TAWSS.

[0428] Instructions FIG. 19 shows a flowchart of a method 1900 for generating training data using a parametric mesh, which method is performed in accordance with one or more non-limiting implementations of the present technology.

[0429] In one or more embodiments, server 230 comprises at least a processor, such as processor 110 and / or GPU 111, operably connected to a non-transitory computer-readable storage medium, such as solid-state drive 120 and / or random access memory 130, that stores computer-readable instructions. The at least one processor is configured to or operable to perform method 1900 upon execution of the computer-readable instructions.

[0430] The method 1900 begins with a process step 1902 .

[0431] According to processing step 1902, at least one processor generates a parametric mesh for a given patient diagnosed with AAA based on a set of baseline images, the parametric mesh encoding features at individual node locations.

[0432] According to process step 1904, at least one processor receives a set of follow-up images for a given patient.

[0433] According to process step 1906, at least one processor generates a follow-up mask on the parametric mesh based on the set of follow-up images.

[0434] According to process step 1908, at least one processor generates mask comparison data based on the selected features and selected regions between the masks of the parametric meshes.

[0435] According to process step 1908, at least one processor associates labels with the parametric mesh and the patient comparison data.

[0436] Processing steps 1902-1908 are repeated for multiple patients.

[0437] FIG. 20 shows a flowchart of a method 2000 for training a model to perform AAA growth prediction based on training data generated from a parametric mesh, the method being performed in accordance with one or more non-limiting implementations of the present technology.

[0438] In one or more embodiments, server 230 comprises at least a processor, such as processor 110 and / or GPU 111, operably connected to a non-transitory computer-readable storage medium, such as solid-state drive 120 and / or random access memory 130, that stores computer-readable instructions. The at least one processor is configured to or operable to perform method 2000 upon execution of the computer-readable instructions.

[0439] According to processing step 2002, at least one processor receives a training data set associated with individual labels, the training data set being generated by comparing features between a baseline mask and a follow-up mask of the individual parametric meshes, the training data set being generated using method 1900.

[0440] According to process step 2004, at least one processor receives a set of growth prediction models 270.

[0441] According to processing step 2006, at least one processor trains a set of growth prediction models 270 on the training data set.

[0442] According to processing step 2008, the at least one processor outputs a set of trained growth prediction models 270.

[0443] FIG. 21 shows a flowchart of a method 2100 for performing AAA growth prediction using a trained model, the method being carried out in accordance with one or more non-limiting implementations of the present technology.

[0444] In one or more embodiments, server 230 comprises at least a processor, such as processor 110 and / or GPU 111, operably connected to a non-transitory computer-readable storage medium, such as solid-state drive 120 and / or random access memory 130, that stores computer-readable instructions. The at least one processor is configured or operable to perform method 2100 upon execution of the computer-readable instructions.

[0445] According to processing step 2100, at least one processor receives a baseline image set of a body including an aorta of a given patient, the image set including at least one image, the image set being acquired using a medical imaging device.

[0446] According to processing step 2102, at least one processor segments the image set using at least one trained segmentation model to obtain segmented regions of interest (ROIs) of the aorta and adjacent structures.

[0447] According to process step 2106, the at least one processor generates wall shear stress parameters based on the segmented ROI of the aorta.

[0448] According to processing step 2108, the at least one processor determines an intraluminal thickness parameter based on the segmented ROI of the aorta.

[0449] According to processing step 2110, at least one processor generates a 3D parametric mesh based on the segmented ROI of the aorta, the 3D parametric mesh including multiple concentric 3D mesh layers, each of the multiple concentric 3D mesh layers including the same predetermined number of nodes, and the generating includes encoding the segmented ROI, wall shear stress parameters, and luminal thickness parameters as features at individual node locations.

[0450] According to processing step 2112, at least one processor predicts whether a given patient will exhibit AAA growth based on at least a subset of the features of the parametric mesh using the trained growth prediction model.

[0451] Preliminary experimental results have been obtained based on one or more non-limiting embodiments of the present technology.

[0452] Preliminary experiment results Of the 36 patients (3,456 patches), 3,147 patches were used for AI modeling. 309 patches, randomly distributed across patients, were excluded due to failure of the diameter growth calculator's quality check, which typically occurs near the bifurcation of the aorta into the iliac arteries. The Extra Tree algorithm was used as a binary classifier, with the positive class representing patches growing >2.5 mm / year. Before training the algorithm, relevant features were selected using Boruta feature selection. To evaluate the algorithm's performance, a stratified 70% / 30% training / test dataset split at the patient level (25 patients for training and 11 patients for testing) was implemented. Training and inference were performed at the patch level within each patient, with the training / test split based on random sampling of patients. Ten-fold cross-validation was performed on the training set, and the 30% set was used as a pure validation set. Therefore, to avoid missing labels, patch samples from a patient were not included in both the training and test sets. The training dataset was used to train the ExtraTree model, and the test dataset was used to evaluate its performance in terms of the area under the ROC curve. All analyses were performed using the Python programming language and the scikit-learn library.

[0453] Additional biomarkers derived from clinical and demographic information, including baseline maximum aortic diameter, age, biological sex, weight, height, family history of AAA, smoking history, heart disease, hypertension (HTN), chronic obstructive pulmonary disease (COPD), and diabetes mellitus (DM), were also investigated as predictors of growth.

[0454] In the AAA study population (n = 36, mean age 77 ± 7 years, 89% male), the mean maximum aortic diameter at baseline was 47.2 ± 5.7 mm, and the median surveillance period between CT scans was 12 months (range 8–31 months). Patient demographic and clinical information is summarized in Table 1.

[0455] [Table 1]

[0456] HTN=hypertension; COPD=chronic obstructive pulmonary disease, DM=diabetes mellitus.

[0457] Of a total of 3147 patches evaluated according to regional diameter growth, 728 patches (23%) showed growth acceleration above the relevant threshold at the time of follow-up evaluation. The maximum growth rate of an individual aorta occurred in only two patients (6%) at the location of the maximum baseline diameter.

[0458] Patients with a larger baseline maximum diameter (≥50 mm) showed no significant differences in regional diameter growth, regional ILT thickness, or regional strain when compared with patients with a smaller baseline maximum diameter (<50 mm).For TAWSS, a significant difference was found between the two subsets, with patients with a larger baseline maximum diameter having a significantly lower regional TAWSS (mean regional TAWSS 0.59±0.37 Pa vs. 0.78±0.48 Pa, p<0.001).

[0459] Among patients in the subset with smaller baseline maximum diameter, patients with faster diameter growth (exceeding the median maximum annual growth rate per patient) had significantly higher local ILT thickness (mean local ILT 4.87 ± 3.37 mm vs. 3.71 ± 2.77 mm, p < 0.001) and significantly lower local TAWSS (mean local TAWSS 0.49 ± 0.38 Pa vs. 0.83 ± 0.48 Pa, p < 0.001). On the other hand, among patients in the subset with larger baseline maximum diameter, patients with faster diameter growth (exceeding the median maximum annual growth rate per patient) had significantly higher local ILT thickness (mean local ILT 5.31 ± 3.57 mm vs. 4.96 ± 3.62 mm, p < 0.001), but no significant differences were observed for local strain or TAWSS.

[0460] The area under the curve (AUC) of the receiver operating characteristic (ROC) curve constructed for the extra-tree classifier was statistically greater than 0.5 (AUC = 0.92, micro-AUC and macro-AUC were 0.94 and 0.92, respectively) (Figure 17), indicating the excellent performance of the model in predicting associated aortic growth.

[0461] Shapley Additive Explanations (SHAP) dependency plots 1610, 1620, 1630, and 1640 are presented to show the contribution and importance of the explored biomarkers to growth prediction (Figure 16). The three biomechanics-based biomarkers (TAWSS, strain, and ILT), which are components of the RW index, were found to be important features contributing to regional growth, with TAWSS playing the most important role in model prediction. Additional clinical biomarkers were found to have less impact on growth prediction.

[0462] Characterization of aortic tissue with biomechanics-based biomarkers according to one or more embodiments of the present technology has demonstrated favorable performance in AI-based prediction of faster-than-average growth in a continuously monitored AAA population. This approach provides functional insight into the multifactorial nature of AAA pathophysiology, accounting for its regional and heterogeneous nature. Functional biomarkers were objectively selected as primary contributors to associated aortic growth.

[0463] Given the ongoing and rapid increase in risk associated with AAA patients, access to information on disease progression is essential for improving disease management. The ability to access functional information related to baseline tissue weakening and disease progression for individual aortas has the potential to benefit patient monitoring, risk stratification, and treatment selection, optimizing precision-based aortic care.

[0464] One or more embodiments of the present technology provide an anatomically relevant meshing strategy to produce homogenized data across multiple modalities and scans. Parametric meshes generated using the present disclosure allow data from all data types, ranging from shell and solid meshes to array-like data including pixel-specific data, to be stored in stackable layers that are easy to interpret and utilize, relying on a single type of data encoding, for training more compact machine learning-based models.

[0465] One or more embodiments of the present method and system convert multiple vessel-specific data types into multi-channel, anatomically relevant, stackable images that can be used to train diagnostic and prognostic artificial intelligence-based models, as well as provide organized, intuitive multi-domain reports to medical professionals.

[0466] One or more embodiments of the present technology enable modular modeling for diagnostic and prognostic purposes, leveraging each available domain of information. Because all models rely on the same data types, weights can be shared or retrained very minimally when new domains of information are introduced. Modular modeling makes it possible to retrain new architectures or for new tasks by utilizing fewer weights (parameters) and requiring less retraining of these weights. Consequently, this makes it easier to obtain generalizable models starting with a smaller number of vascular scans.

[0467] In some cases, what are believed to be useful examples of modifications to the technology may also be described. This is done merely to aid in understanding and, again, does not define the scope or delimit the technology. These modifications are not an exhaustive list, and one of ordinary skill in the art may nonetheless make other modifications while remaining within the scope of the technology. Furthermore, if examples of modifications are not described, it should not be construed that modifications are not possible and / or that what is described is the only way to implement that element of the technology.

[0468] Modifications and improvements to the above-described embodiments of the technology may become apparent to those skilled in the art. The foregoing description is intended to be illustrative, not limiting.

Claims

1. 1. A method for predicting abdominal aortic aneurysm (AAA) growth based on at least one image of a given patient previously diagnosed with AAA, the method being executed by at least one processor, the at least one processor having access to a trained growth prediction machine learning (ML) model, the method comprising: receiving a baseline image set of the given patient's body including the aorta, the image set including at least one image, the image set being acquired using a medical imaging device; segmenting the image set using at least one trained segmentation model to obtain segmented regions of interest (ROIs) of the aorta and adjacent structures; generating a wall shear stress parameter based on the segmented ROI of the aorta; determining a luminal thickness parameter based on the segmented ROI of the aorta; generating a 3D parametric mesh based on the segmented ROI of the aorta, the 3D parametric mesh including a plurality of concentric 3D mesh layers, each of the plurality of concentric 3D mesh layers including the same predetermined number of nodes, the generating including encoding the segmented ROI, the wall shear stress parameter, and the luminal thickness parameter as features at individual node locations; predicting whether the given patient will exhibit AAA growth using a trained growth prediction ML model based on at least a subset of features of the 3D parametric mesh; A method comprising:

2. The method of claim 1 , wherein the ROI of the aorta and adjacent structures includes the abdominal aortic region and the iliac arteries.

3. The method of claim 2 , wherein the ROI of the aorta and adjacent structures includes a portion of the spine.

4. 4. The method of claim 1, wherein generating the 3D parametric mesh based on the segmented ROI of the aorta comprises encoding pixel locations and pixel intensity values ​​at the individual node locations.

5. The method of any one of claims 1 to 4, wherein the subset of features comprises geometric features, the geometric features comprising 2D distances to a centerline of the parametric mesh.

6. The method of claim 5 , wherein the geometric features include 3D distances to a centerline of the parametric mesh.

7. prior to receiving the baseline image set; and obtaining the trained growth prediction model by training the growth prediction model with a training dataset, the training dataset comprising, for each individual patient of a plurality of patients: an individual comparison of encoded features between a baseline mask and a follow-up mask of an individual 3D parametric mesh generated for said individual patient based on an individual set of baseline images and an individual set of follow-up images; a separate growth label indicating the presence of AAA growth; The method according to any one of claims 1 to 6, comprising:

8. The method of claim 7 , wherein the individual comparison of encoded features comprises a comparison of diameters at individual sections of nodes perpendicular to individual aortic centerlines of the individual parametric meshes.

9. 9. The method of claim 7 or 8, wherein the individual comparison of encoded features comprises a comparison of diameters at individual sections of nodes perpendicular to individual aortic centerlines of the individual parametric meshes.

10. The method of any one of claims 1 to 9, wherein the trained growth prediction model comprises a decision tree.

11. 1. A method for predicting growth of an aneurysm in a blood vessel based on at least one image of a given patient previously diagnosed with said aneurysm, the method being executed by at least one processor, the at least one processor having access to a trained growth prediction machine learning (ML) model, the method comprising: receiving a segmented region of interest (ROI) of the vessel and adjacent structures segmented from a set of images of the given patient acquired using a medical imaging device; generating a wall shear stress parameter based on the segmented ROI of the blood vessel; determining an endoluminal thickness parameter based on the segmented ROI of the blood vessel; generating a 3D parametric mesh based on the segmented ROI of the blood vessel, the 3D parametric mesh including a plurality of concentric 3D mesh layers, each of the plurality of concentric 3D mesh layers including the same predetermined number of nodes, the generating including encoding the segmented ROI, the wall shear stress parameter, and the luminal thickness parameter as features at individual node locations; predicting whether the given patient will exhibit aneurysm growth using the trained growth prediction ML model based on at least a subset of features of the 3D parametric mesh; A method comprising:

12. 1. A system for predicting abdominal aortic aneurysm (AAA) growth based on at least one image of a given patient previously diagnosed with AAA, the system comprising: a non-transitory storage medium storing computer-readable instructions; at least one processor operatively connected to the non-transitory storage medium; the at least one processor has access to a trained growth predictive machine learning (ML) model; The at least one processor, when executing the computer-readable instructions, receiving a baseline image set of the given patient's body including the aorta, the image set including at least one image, the image set being acquired using a medical imaging device; segmenting the image set using at least one trained segmentation model to obtain segmented regions of interest (ROIs) of the aorta and adjacent structures; generating a wall shear stress parameter based on the segmented ROI of the aorta; determining a luminal thickness parameter based on the segmented ROI of the aorta; generating a 3D parametric mesh based on the segmented ROI of the aorta, the 3D parametric mesh including a plurality of concentric 3D mesh layers, each of the plurality of concentric 3D mesh layers including the same predetermined number of nodes, the generating including encoding the segmented ROI, the wall shear stress parameter, and the luminal thickness parameter as features at individual node locations; predicting whether the given patient will exhibit AAA growth using a trained growth prediction ML model based on at least a subset of features of the 3D parametric mesh; A system that is configured to:

13. The system of claim 12 , wherein the ROI of the aorta and adjacent structures includes the abdominal aortic region and the iliac arteries.

14. The system of claim 13 , wherein the ROI of the aorta and adjacent structures includes a portion of the spine.

15. 15. The system of claim 12, wherein generating the parametric mesh based on the segmented ROI of the aorta comprises encoding pixel locations and pixel intensity values ​​at the individual node locations.

16. The system of any one of claims 12 to 15, wherein the subset of features comprises geometric features, the geometric features comprising 2D distances to a centerline of the parametric mesh.

17. The system of claim 16 , wherein the geometric features include a 3D distance to a centerline of the parametric mesh.

18. The at least one processor may further, prior to receiving the baseline image set: and obtaining the trained growth prediction model by training the growth prediction model with a training dataset, the training dataset comprising, for each individual patient of a plurality of patients: an individual comparison of encoded features between a baseline mask and a follow-up mask of an individual 3D parametric mesh generated for said individual patient based on an individual set of baseline images and an individual set of follow-up images; a separate growth label indicating the presence of AAA growth; The system according to any one of claims 12 to 17, comprising:

19. 20. The system of claim 18, wherein the individual comparison of encoded features comprises a comparison of diameters at individual sections of nodes perpendicular to individual aortic centerlines of the individual parametric meshes.

20. 20. The system of claim 18 or 19, wherein the individual comparison of encoded features comprises a comparison of diameters at individual sections of nodes perpendicular to individual aortic centerlines of the individual parametric meshes.

21. The system of any one of claims 12 to 20, wherein the trained growth prediction model comprises a decision tree.

22. 1. A system for predicting aneurysm growth based on at least one image of a given patient previously diagnosed with said aneurysm, the system comprising: a non-transitory storage medium storing computer-readable instructions; at least one processor operatively connected to the non-transitory storage medium; The at least one processor has access to a trained growth prediction machine learning (ML) model, and the at least one processor, when executing the computer-readable instructions, receiving a segmented region of interest (ROI) of the vessel and adjacent structures segmented from a set of images of the given patient acquired using a medical imaging device; generating a wall shear stress parameter based on the segmented ROI of the blood vessel; determining an endoluminal thickness parameter based on the segmented ROI of the blood vessel; generating a 3D parametric mesh based on the segmented ROI of the blood vessel, the 3D parametric mesh including a plurality of concentric 3D mesh layers, each of the plurality of concentric 3D mesh layers including the same predetermined number of nodes, the generating including encoding the segmented ROI, the wall shear stress parameter, and the luminal thickness parameter as features at individual node locations; predicting whether the given patient will exhibit aneurysm growth using the trained growth prediction ML model based on at least a subset of features of the 3D parametric mesh; A system that is configured to:

Images