Anatomical and functional assessment of coronary artery disease using machine learning

Abstract

Anatomical and functional assessment of coronary artery disease (CAD) using machine learning and computational modeling techniques deploys methodologies for noninvasive fractional flow reserve (FFR) quantification based on angiography-derived anatomical and hemodynamic data, relying on machine learning algorithms for image segmentation and blood flow assessment, and accurate physics-based computational fluid dynamics (CFD) simulations for the calculation of FFR.

Description

[Technical field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority to U.S. Provisional Patent Application No. 62 / 939,370, entitled "Anatomical and Functional Assessment of CAD Using Machine Learning," filed November 22, 2019, the entire disclosure of which is expressly incorporated herein by reference.

[0002] The present invention relates generally to fully automated detection of coronary vessels and their branches in angiograms, and more specifically to calculation of the diameter of such vessels, detection of stenosis, and determination of the percentage narrowing of a stenosis and functional blood flow limitation therefrom. [Background technology]

[0003] The description of the background art provided herein is for the purpose of generally presenting the context of the present disclosure. The work of the presently named inventors to the extent described in this Background section, and aspects of the description that may not otherwise be considered prior art at the time of filing, are not admitted, expressly or impliedly, as prior art to the present disclosure.

[0004] Coronary artery disease (CAD) is one of the leading causes of death in the United States, affecting more than 15 million Americans. CAD is characterized by the buildup of atherosclerotic plaque in the coronary arteries, which can lead to narrowing (also known as stenosis) or blockage of the coronary arteries, resulting in symptoms such as angina and potentially myocardial infarction.

[0005] As used herein, a "vascular occlusion condition" refers to what is commonly understood as CAD, e.g., narrowing (localized or diffuse) of the epicardial coronary arteries as seen on imaging studies and characterized by either anatomical or functional indicators. Conversely, a "microvascular disease condition" refers to a disease of the coronary microcirculation characterized by a loss of vasodilatory capacity.

[0006] The primary invasive diagnostic method for CAD is coronary angiography, in which a contrast agent is injected into a patient's blood vessels via a catheter and imaged to characterize the severity of stenosis. This method relies on visualization of anatomical abnormalities and is semi-quantitative, as visual inspection merely estimates the percentage of luminal area reduction. Diameter reductions estimated at 70% or greater often result in further evaluation or revascularization procedures, such as coronary stent placement.

[0007] A more quantitative, physiology-based approach to assessing coronary stenosis is to calculate fractional flow reserve (FFR), a metric defined as the ratio of hyperemic blood flow in a diseased artery to the expected hyperemic blood flow in the same artery without stenosis. For example, in coronary arteries, FFR can be expressed as the ratio of distal coronary artery pressure to proximal coronary artery pressure. For example, an FFR less than 0.80 indicates the presence of severe stenosis requiring revascularization due to compromised blood flow to vascular beds distal to the vessel. Revascularization decisions incorporating FFR have been shown to improve outcomes compared to angiography alone. FFR determination is also robust to various patient geometries, taking into account, for example, the contribution of collateral vessels to blood flow and lesion geometry.

[0008] Despite its advantages, medical professionals often do not measure FFR in patients due to the invasive nature of using catheter-based pressure measurements. Some have found that physicians choose not to perform FFR in nearly two-thirds of all cases, citing risk to the patient, lack of resources, and additional costs. Another drawback of FFR is its variability due to different hemodynamic conditions within a patient.

[0009] There is a need for more accurate and less invasive techniques for diagnosing CAD. More specifically, there is a need for a non-invasive and user-independent approach to measuring FFR that would not pose a risk to the patient.

[0010] Recently, others have proposed non-invasive computational workflows to determine FFR and assess the severity of CAD. These efforts have adopted two different approaches. The first approach relies on computed tomography angiography (CTA) data to reconstruct the 3D geometry of the coronary artery tree. The vessel lumen is segmented using edge detection or machine learning algorithms, and the output is manually corrected by an expert. Computational fluid dynamics (CFD) simulations are performed on the vessel geometry derived from CTA, and the FFR is calculated on the company's server or cloud. Although this non-invasive approach has shown promising results in clinical trials, the use of CTA is heavily affected by imaging artifacts due to calcification, making it difficult to delineate the vessel lumen border. CTA also has limited ability to detect small continuous lesions or capture fine vessel geometry due to its lower spatial resolution compared to angiography. Finally, because CTA data do not provide information on blood flow, boundary conditions (BC) for CFD analysis of hemodynamics typically rely on morphometric or population data and are therefore not patient-specific.

[0011] The second approach, also based on a non-invasive computational workflow, relies on multi-plane angiography data to reconstruct vascular geometry before performing physics-based blood flow simulations. The main advantages of using angiography are its superior ability to detect vessel lumen boundaries in the presence of calcified stenoses and its high spatial resolution compared to CTA, improving the sensitivity of geometry reconstruction algorithms. In addition, time-resolved angiography has information about how contrast moves within the vessel and can therefore be used to estimate blood flow velocities and inform BC for CFD analysis. However, angiography-based approaches for FFR quantification have a fundamental challenge: the reconstruction of the 3D geometry of the vessel of interest from a series of 2D images acquired at various positions on the patient table over time. Furthermore, all angiography-based approaches for FFR quantification created workflows that required substantial input by the operator to identify the vessel of interest and aid in the segmentation. Finally, all angiography-based approaches either considered the reconstruction of a single coronary artery or modeled the physics of blood flow using highly simplified methods. These drawbacks effectively negate the benefits of using high-resolution angiography data, the most commonly performed procedure for CAD diagnosis.

[0012] Regardless of the approach, all computationally derived FFR methods have shown poor predictive performance around the critical diagnostic value of FFR=0.8 due to the aforementioned limitations in image data quality, lack of information on blood flow, need for operator input, and computational modeling assumptions. Thus, there is currently no pipeline for accurate FFR calculation that can be effectively deployed in every hospital across the country.

[0013] There is a significant need for more accurate, minimally invasive techniques for CAD diagnosis that use fully automated, non-invasive approaches for FFR measurement. Summary of the Invention

[0014] Techniques are provided for anatomical and functional assessment of coronary artery disease (CAD) using machine learning and computational modeling techniques. The techniques herein address shortcomings of traditional CAD assessment by using a novel methodology for noninvasive fractional flow reserve (FFR) quantification based on angiography-derived anatomical and hemodynamic data, relying on machine learning algorithms for image segmentation and blood flow assessment, and relying on accurate physics-based computational fluid dynamics (CFD) simulations for the calculation of FFR.

[0015] In an exemplary embodiment, the technology provides a process for assessing both the anatomical and functional severity of CAD through customized machine learning and computational modeling methods from the use of both static and dynamic coronary angiography data, with the end goal of determining FFR in a less risky and more accurate manner.

[0016] The use of functional markers of disease has gained significant traction in the past few years, replacing simpler anatomical structure-based markers. In the field of cardiology, fractional flow reserve (FFR) is a hemodynamic index (i.e., normalized pressure gradient under conditions of maximum blood flow) that has demonstrated better diagnostic outcomes than anatomical-based markers. In exemplary embodiments herein, two-dimensional (2D) angiographic data is used, specifically dynamic angiographic data that provides a description of the transport of contrast agent (dye) down the vessel of interest. The 2D time-resolved angiographic data is used to inform computational simulations, thereby obtaining more accurate predictions of FFR than would be the case in the absence of dynamic transport of contrast agent down the vessel. Additionally, in exemplary embodiments, three-dimensional (3D) geometric models are generated from the 2D angiographic data, and those 3D geometric models are used to simulate hemodynamic conditions, including FFR.

[0017] According to one example, a computer-implemented method for assessing CAD includes receiving, by one or more processors, angiographic image data of a vascular examination area of ​​a subject, the angiographic image data including an angiographic image captured over a sampling period; applying, by the one or more processors, the angiographic image data to a vascular segmentation machine learning model and generating, by the one or more processors, a two-dimensional (2D) segmented vascular image of the vascular examination area using the vascular segmentation machine learning model; generating, by the one or more processors, a three-dimensional (3D) segmented vascular tree geometric model of a blood vessel in the vascular examination area from the 2D segmented vascular image; applying, by the one or more processors, the 3D segmented vascular tree geometric model to a fluid dynamics machine learning model to assimilate blood flow data over the sampling period of one or more blood vessels in the vascular examination area using the fluid dynamics machine learning model; applying, by the one or more processors, the 3D segmented vascular tree geometric model and the assimilated blood flow data to a 3D high-fidelity computational fluid dynamics model; and determining, by the one or more processors, a state of vascular occlusion of one or more blood vessels in the vascular examination area.

[0018] According to one example, a computing device configured to assess CAD includes one or more processors and, when executed, the one or more processors include receiving, by the one or more processors, angiographic image data of a vascular examination region of a subject, the angiographic image data including angiographic images captured over a sampling period; applying, by the one or more processors, the angiographic image data to a vascular segmentation machine learning model and generating, using the vascular segmentation machine learning model, a two-dimensional (2D) segmented vascular image of the vascular examination region; and generating, by the one or more processors, a vascular segmentation image from the 2D segmented vascular image. The system includes one or more computer-readable memories storing instructions that cause the one or more processors to: generate a three-dimensional (3D) segmented vascular tree geometric model of a blood vessel within a vascular examination area; apply, by the one or more processors, the 3D segmented vascular tree geometric model to a fluid dynamics machine learning model to assimilate blood flow data over a sampling period of one or more blood vessels within the vascular examination area using the fluid dynamics machine learning model; apply, by the one or more processors, the 3D segmented vascular tree geometric model and the assimilated blood flow data to the 3D high-fidelity computational fluid dynamics model; and determine, by the one or more processors, a state of vascular occlusion of one or more blood vessels within the vascular examination area.

[0019] According to another example, a computer-implemented method for assessing coronary artery disease includes receiving, by one or more processors, a plurality of angiographic images of a vascular examination area of ​​a subject, the angiographic images being captured over a sampling period, the vascular examination area including one or more blood vessels; applying, by the one or more processors, the angiographic images to a vascular segmentation machine learning model and generating, by the one or more processors, two-dimensional (2D) segmented vascular images of the one or more blood vessels using the vascular segmentation machine learning model; generating, by the one or more processors, a one-dimensional (1D) segmented vascular tree geometric model of the one or more blood vessels from the 2D segmented vascular images; applying, by the one or more processors, the 1D segmented vascular tree model to a fluid dynamics machine learning model to assimilate blood flow data over the sampling period of the one or more blood vessels using the fluid dynamics machine learning model; applying, by the one or more processors, the 1D segmented vascular tree model and the assimilated blood flow data to a graph theory-based reduced dimensional model based on a computational fluid dynamics model; and determining, by the one or more processors, a state of vascular occlusion.

[0020] A computing device configured to assess CAD includes one or more processors, and when executed, the computing device includes the one or more processors: receiving, by the one or more processors, a plurality of angiographic image data of a vascular examination region of a subject, the angiographic images being captured over a sampling period, the vascular examination region including one or more blood vessels; applying, by the one or more processors, the angiographic images to a vascular segmentation machine learning model and generating, using the vascular segmentation machine learning model, two-dimensional (2D) segmented vascular images of the one or more blood vessels; and one or more computer-readable memories storing instructions that cause the one or more processors to: generate a one-dimensional (1D) segmented vascular tree geometric model of one or more blood vessels from the computed vascular images; apply, by the one or more processors, the 1D segmented vascular tree model to a fluid dynamics machine learning model to assimilate blood flow data over a sampling period of the one or more blood vessels using the fluid dynamics machine learning model; apply, by the one or more processors, the 1D segmented vascular tree model and the assimilated blood flow data to a graph theory based reduced dimensional model that is based on a computational fluid dynamics model; and determine, by the one or more processors, a state of vascular occlusion. [Brief description of the drawings]

[0021] The drawings described below depict various aspects of the systems and methods disclosed herein. It is understood that each figure depicts an embodiment of a particular aspect of the disclosed systems and methods, and each of the figures is intended to correspond to a possible embodiment thereof. Furthermore, wherever possible, the following description will refer to reference numerals contained in the following figures, and features shown in multiple figures will be designated with consistent reference numerals.

[0022] [Figure 1] FIG. 1 is a schematic diagram of an exemplary system for performing anatomical and functional assessment of coronary artery disease (CAD) using a machine learning framework for evaluating angiographic images, according to one example. [Diagram 2] 2 is a schematic diagram of a CAD assessment machine learning and computational modeling framework, such as may be implemented in the system of FIG. 1, according to an example. [Diagram 3] FIG. 1 is a process flow diagram of a method for assessing coronary artery disease, according to an example. [Figure 4] FIG. 4 is a process flow diagram of a method for generating two-dimensional (2D) segmented vascular images from angiography image data, as may be performed by a vascular segmentation machine learning model as part of the process of FIG. 3, according to one example. [Diagram 5] 5 illustrates an exemplary configuration of the vessel segmentation machine learning model of FIG. 4. [Figure 6] 5 is a process flow diagram of a series of exemplary methods for generating a three-dimensional (3D) segmented vascular tree model from a 2D segmented vascular image, according to one example, which generated the process of FIG. 4 . [Figure 7] An example illustrates a process for defining a graph-theoretic based reduced dimensional model that can accurately and efficiently characterize the functional state of a CAD. [Figure 8] 1 illustrates an image of a clinical angiogram image that may be used to train or predict a vessel segmentation machine learning model and a fluid dynamics machine learning model, according to an example. [Figure 9] 1 illustrates an image of a synthetic angiography image that may be used to train a vessel segmentation machine learning model and a fluid dynamics machine learning model, according to an example. [Figure 10A] 1 illustrates an exemplary 3D vascular tree model and various lumped parameter models that may be used by a computational fluid dynamics system model for each of their parameters. [Figure 10B] 1 illustrates an exemplary 3D vascular tree model and various lumped parameter models that may be used by a computational fluid dynamics system model for each of their parameters. [Figure 11-1] 1 illustrates an exemplary process for generating an anatomical and functional assessment of CAD, according to an example. [Figure 11-2]1 illustrates an exemplary process for generating an anatomical and functional assessment of CAD, according to an example. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0023] Techniques are provided for performing anatomical and functional assessment of coronary artery disease (CAD) by analyzing dynamic angiographic image data using physics-based machine learning from neural networks, order reduction of high-fidelity models using graph theory, together with machine learning based anatomical segmentation and high-fidelity computer simulation of hemodynamics to create an automated workflow that can provide superior diagnostic performance for CAD. Techniques are provided for performing noninvasive fractional flow reserve (FFR) quantification based on angiographic data, relying on machine learning algorithms for image segmentation, physics-based machine learning and computational fluid dynamics (CFD) simulations for more accurate functional assessment of blood vessels. Specifically, in some examples, two-dimensional dynamic angiographic data is used to capture transport of dye down the blood vessel of interest. This dynamic information can be used to inform computational simulations and therefore to take more accurate predictions of FFR, especially in borderline cases. Furthermore, angiographic data does not provide three-dimensional anatomical information, but current techniques include processes for developing image reconstruction algorithms to obtain three-dimensional (3D) and one-dimensional (1D) geometric models of a patient's vasculature that are then used for computer simulation of hemodynamics. Although techniques are described herein with respect to determining FFR, the same techniques may be used to calculate other fractional flow reserve metrics. Thus, references to determining FFR in the examples herein include determining instantaneous fractional flow reserve (iFR), quantitative coronary flow ratio (QFR), and the like.

[0024] In some examples, systems and methods are provided for assessing coronary artery disease. The system may receive angiographic image data of a vascular examination region of a subject. The angiographic image data may contain a plurality of angiographic images captured over a sampling period. The system may apply the angiographic image data to a first machine learning model, a vascular segmentation machine learning model. The vascular segmentation machine learning model may generate two-dimensional (2D) segmented vascular images of the vascular examination region, and from these 2D segmented vascular images, a 3D geometric vascular tree model is generated that models the vessels having the vascular examination region. In other examples, a 1D equivalent vascular tree model may be generated from the 3D vascular tree model. The 3D or 1D geometric vascular tree model may be applied to a second machine learning model, a fluid dynamics machine learning model, to assimilate blood flow data over a sampling period of one or more vessels in the vascular examination region. From the assimilated blood flow data and from the 3D or 1D geometric vascular tree model, the computational fluid dynamics model is configured to determine a state of blood vessels in the vasculature, which may include a state of vascular occlusion and / or a state of microvascular disease / resistance. Specifically, to determine microvascular disease / resistance, angiographic images may be acquired under two different hemodynamic conditions, one at a baseline condition and one at a hyperemic (high flow) condition, and compared between the two. In yet another example, the microvasculature may be assessed solely from examining angiographic images captured during the hyperemic condition.

[0025] In FIG. 1, CAD assessment system 100 includes a computing device 102 (or “signal processor” or “diagnostic device”) configured to collect angiographic image data from a patient 120 via an angiographic imaging device 124 by performing functions of the disclosed embodiments. As shown, system 100 may be implemented on computing device 102, specifically on one or more processing units 104, which may represent central processing units (CPUs), and / or one or more graphics processing units (GPUs), including clusters of CPUs and / or GPUs, any of which may be cloud-based. Features and functions described for system 100 may be stored on and implemented from one or more non-transitory computer-readable media 106 of computing device 102. Computer-readable media 106 may include, for example, an operating system 108 and a CAD machine learning (deep learning) framework 110 having elements corresponding to elements of the deep learning framework described herein. More generally, the computer-readable medium 106 may store trained deep learning models, including vascular segmentation machine learning models, fluid dynamics machine learning models, graph theory based reduced dimensional models, executable code, etc., used to implement the techniques herein. The computer-readable medium 106 and processing unit 104 may store image data, segmentation models or rules, fluid dynamic classifiers, and other data herein in one or more databases 112. As discussed in the examples herein, the CAD machine learning framework 110 applying the techniques and processes herein (e.g., various different neural networks) may generate 3D and / or 1D segmented vascular tree geometric models, FFR and other fluid dynamic assessments, status data of vascular occlusion, and / or microvascular disease data.

[0026] The computing device 102 includes a network interface 114 communicatively coupled to a network 116 for communication to and / or from a portable personal computer, a smartphone, an electronic document, a tablet, and / or a desktop personal computer, or other computing device. The computing device further includes an I / O interface 115 connected to devices such as a digital display 118, a user input device 122, and the like. As described herein, the computing device 102 generates a representation of the subject's CAD, which may include the status of blood vessels in the vasculature, such as the status of vascular occlusion (anatomically and functionally via FFR calculation, via iFR calculation, or via QFR calculation) and the status of microvascular disease prediction (by contrasting the change in peripheral resistance when the two hemodynamic states are recorded), as an electronic document that can be accessed and / or shared over the network 116.

[0027] In the illustrated example, the computing device 102 is communicatively coupled to an electronic medical record (EMR) database 126 via the network 116. The EMR 126 may be a network-accessible database or a dedicated processing system. In some examples, the EMR 126 includes data for one or more respective patients. The EMR data may include vital signs data (e.g., hemoglobin oxygen saturation derived from pulse oximetry, heart rate, blood pressure, respiratory rate), lab data such as a complete blood count (e.g., mean platelet volume, hematocrit, hemoglobin, mean red blood cell hemoglobin, mean red blood cell concentration, mean red blood cell volume, white blood cell count, platelets, red blood cell count, and red blood cell distribution width), lab data such as a basic metabolic panel (e.g., blood urea nitrogen, potassium, sodium, glucose, chloride, CO2, calcium, chloride, potassium, sodium, glucose, chloride, potassium ... atinine), demographic data (e.g., age, weight, race and sex, zip code), less common laboratory data (e.g., bilirubin, partial thromboplastin time, international normalized ratio, lactate, magnesium and phosphorous), and other suitable patient indices now existing or hereafter developed (e.g., O2, use of Glasgow Coma Score or components thereof, and urine volume in the past 24 hours, antibiotic administration, blood transfusion, fluid administration, etc.), as well as calculations including shock index and mean arterial pressure. The EMR data may additionally or alternatively include chronic medical and / or surgical conditions. The EMR data may include historical data collected from previous examinations of the patient, including historical FFR, iFR, or QFR data. Previous determinations of stenosis, predictions of vascular disease, vascular resistance, CFD simulation data, and other data generated by the techniques herein. The EMR 126 may be updated as new data is collected from the angiography imaging device 124 and evaluated using the computing device 102. In some examples, the techniques may provide for ongoing training of the EMR 126.

[0028] In conventional angiographic imaging applications, angiographic images are captured by a medical imaging device and then sent to an EMR for storage and, in some instances, further processing, including image processing, before those images are sent to a medical professional. In the present technology, the status of occlusion and the status of microvascular disease can be determined at a computing device based on the angiographic images without first offloading those images to the EMR 126 for processing. Overall, the technology proposed herein can significantly reduce the cardiologist's analysis time, in part due to this bypassing of the EMR 126 for processing. The EMR 126 may simply be polled for data during analysis by the computing device 102, or may be used for storage of status determinations and other calculations generated by the technology herein. Indeed, there are many benefits that arise from faster and more automated analysis resulting from current technology. For example, modeling and vascular occlusion / disease status analysis can be performed separately and sequentially for vessels corresponding to either the left or right coronary tree, but the cardiologist's results are still produced in minutes using, for example, a 3D modeler or a 1D modeler as described herein.

[0029] In the illustrated example, the system 100 is implemented on a single server. However, the functionality of the system 100 may be implemented across distributed devices connected to each other via communication links. In other examples, the functionality of the system 100 may be distributed across any number of devices, including the illustrated portable personal computers, smartphones, electronic documents, tablets, and desktop personal computing devices. In other examples, the functionality of the system 100 may be cloud-based, such as one or more connected cloud CPUs or computing systems, labeled 105, customized to execute the machine learning processes and computational techniques described herein. The network 116 may be a public network, such as the Internet, a private network, such as a research institute or corporate private network, or any combination thereof. The network may include a local area network (LAN), a wide area network (WAN), cellular, satellite, or other network infrastructure, whether wireless or wired. The network may utilize communication protocols, including packet-based and / or datagram-based protocols, such as the Internet Protocol (IP), Transmission Control Protocol (TCP), User Datagram Protocol (UDP), or other types of protocols. Additionally, network 116 may include a number of devices that facilitate network communications and / or form the hardware infrastructure of the network, such as switches, routers, gateways, access points (such as wireless access points as shown), firewalls, base stations, repeaters, backbone devices, etc.

[0030] The computer-readable medium 106 may include executable computer-readable code stored thereon for programming a computer (including, for example, a processor and a GPU) to the techniques herein. Examples of such computer-readable storage media include hard disks, CD-ROMs, digital versatile disks (DVDs), optical storage devices, magnetic storage devices, ROMs (read-only memories), PROMs (programmable read-only memories), EPROMs (erasable programmable read-only memories), EEPROMs (electrically erasable programmable read-only memories), and flash memories. More generally, the processing unit of the computing device 102 may represent a CPU-type processing unit, a GPU-type processing unit, a field programmable gate array (FPGA), another class of digital signal processor (DSP), or other hardware logic components that can be driven by a CPU.

[0031] Although the exemplary deep learning framework herein has been described as being comprised of an exemplary machine learning architecture, it should be noted that any number of suitable convolutional neural network architectures may be used. Generally speaking, the deep learning framework herein may implement any suitable statistical model (e.g., a neural network or other model implemented through a machine learning process) that is applied to each of the received images. As discussed herein, the statistical model may be implemented in a wide variety of ways. In some examples, the machine learning model has the form of a neural network, a support vector machine (SVM), or other machine learning process, and is trained using images or multidimensional data sets to develop a model of vessel segmentation or fluid dynamics calculations. Once these models are properly trained on a series of training images, the statistical model may be used in real time to determine the presence of CAD and to analyze subsequent angiographic image data that is provided as input to the statistical model to determine vascular occlusion status and disease. In some examples, when the statistical model is implemented using a neural network, the neural network may be configured in a wide variety of ways. In some examples, the neural network may be a deep neural network and / or a convolutional neural network. In some examples, the neural network may be a distributed and scalable neural network. The neural network may be customized in a wide variety of ways, such as providing a specific top layer, such as a logistic regression top layer. A convolutional neural network may be viewed as a neural network that includes a set of nodes with associated parameters. A deep convolutional neural network may be viewed as having a structure in which multiple layers are stacked. Neural networks or other machine learning processes may include various sizes, numbers of layers, and levels of connectivity. Some layers may correspond to stacked convolutional layers (optionally followed by contrast normalization and max pooling), followed by one or more fully connected layers.The technique may be implemented such that machine learning training may be performed using small data sets, e.g., less than 10,000 images, less than 1000 images, or less than 500 images. In one example, about 400 images were used. To avoid overfitting, a multi-fold cross-validation process may be used (e.g., 5-fold cross-validation). In some examples, a regularization process such as L1 or L2 may be used to avoid overfitting. For neural networks trained with large data sets, e.g., more than 10,000 images, the number of layers and the size of the layers may be increased by using dropout to address the potential problem of overfitting. In some cases, neural networks may be designed to refrain from using fully connected upper layers at the top of the network. By forcing the network to reduce the dimensionality of the intermediate layers, very deep neural network models may be designed while dramatically reducing the number of learned parameters.

[0032] 2 illustrates an exemplary deep learning framework 200 that is an example of the CAD machine learning framework 110 of the computing device 102. The framework 200 includes a 3D / 1D segmented vascular tree geometric model generator 204 that includes a pre-processor stage 205, a vascular segmentation machine learning model 206 that includes two neural networks, an angiogram processing neural network (APN) 202 and a second stage, and a semantic neural network 207.

[0033] In the illustrated example, the pre-processor 205 receives a clinical angiogram image 203A along with data regarding the contrast injection used to form it. Optionally, the pre-processor 205 may be coupled to receive a synthetic angiogram image 203B, for example, for machine learning training. Additionally, the pre-processor 205 may be coupled to receive a geometrically adjusted vessel image 203C. In some examples, these inputs may be fed directly to the vessel segmentation machine learning model 206, and more specifically, to the APN 202. The pre-processor 205 may perform various pre-processing on the received image data, which may include a noise removal process, a linear filtering process, an image size normalization process, and a pixel intensity normalization process on the received image data.

[0034] The deep learning framework 200 may operate in two different modes, a machine learning training mode and an analysis mode. In the machine learning training mode of the framework, angiographic image data 203A, synthetic angiographic image data 203B, and / or geometrically adjusted (such as horizontal or vertical flip, any level of zoom, rotation, or shear) angiographic image data 203C may be provided to the APN 202. Depending on the data type and data source, different pre-processing functions and values ​​can be applied to the received image data. In the analysis mode, where the machine learning model is trained, the captured angiographic image data 203A of the subject is provided to the APN 202 for analysis and CAD determination. In either mode, the pre-processed image data is provided to a 3D / 1D segmented vascular tree geometric model generator 204, which includes a segmentation machine learning model 206 that receives the pre-processed image data and performs processing in the APN 202 and semantic NN 207, and in the analysis model, generates 2D segmented vascular images. Thus, the vascular segmentation machine learning model 206 may be a convolutional neural network, such as two different convolutional neural networks in a hierarchical configuration as shown in the example of Figure 5. Thus, in some examples, the semantic NN 207 is configured as a modified or unmodified Deeplabv3+ architecture.

[0035] The 3D / 1D segmented vascular tree geometric model generator 204 further includes a 3D modeler 208 configured to generate a 3D vascular tree geometric model of the target area based on the 2D segmented vascular image.

[0036] Once the 3D vascular tree model is generated, the generator 204 may apply further smoothing algorithms and / or surface spline approximation algorithms to further improve the 3D vascular tree model for 3D (e.g., high-fidelity) hemodynamic classification and occlusion analysis.

[0037] To increase processing time and analysis of the state of vascular occlusion of larger vessels and the state of microvascular disease of smaller vessels, in some examples, the techniques herein are implemented with a reduced dimensional model. In some examples, the 3D segmented vascular tree geometric model generated from the captured 2D angiographic images is further reduced to generate a 1D segmented vascular tree geometric model, while still maintaining sufficient data to provide for FFR, iFR, or QFR determination, and computational fluid dynamics modeling. To implement model order reduction, in some examples, the vascular tree geometric model generator 204 includes a 1D modeler 209. The 1D modeler 209 creates a skeletonization of the 3D segmented vascular tree model given by the 3D spatial paths / centerlines of the vessels included in the 3D segmented vascular tree model, and a series of 2D cross-sectional contours separated at any distance along each path / centerline of the tree. An exemplary 1D segmented vascular geometric tree model generated from the 3D segmented vascular geometric tree model is shown in FIG. 6.

[0038] The 3D or 1D vascular tree geometric model from generators 208 or 209 is provided to a blood flow data generator 210 that includes a fluid dynamics machine learning model 212, which may include at least one of the following types of networks: Convolutional Neural Network (CNN), autoencoder, or long short-term memory (LSTM), or a low-dimensional model of blood flow and pressure based on graph theory.

[0039] As shown, the fluid dynamics machine learning model 212 may include many different types of models, both trained and untrained. In some examples, the fluid dynamics machine learning model 212 is a Navier-Stokes informed deep learning framework configured to determine pressure and velocity data across a 3D or 1D vessel space depending on the modeler providing the input 208 or 209. In some examples, the Navier-Stokes informed deep learning framework includes one or more of the following types of methods: Kalman filtering, physically informed neural networks, iterative assimilation algorithms based on contrast arrival time to anatomical landmarks, and TIMI frame counting. Dynamic data regarding a series of images describing the transport of dye down the vessel of interest (see, e.g., FIG. 8) is used to assimilate information regarding blood flow velocity. In some cases, a clinical angiogram may be acquired using a contrast injection system 203D with precise information regarding the pressure applied to the dye bolus, the amount of dye, timing, etc., to provide additional information for the fluid dynamics machine learning model, as also shown.

[0040] In other examples, the fluid dynamics machine learning model 212 includes a graph theory based low dimensional model obtained through a graph of discrepancies between ground truth data (in silico or clinical) including geometry, pressure, blood flow, and indices such as FFR, iFR, or QFR. Any of the techniques herein for defining a graph theory based low dimensional model can generate faster results compared to occlusion analysis techniques based on finite element modeling (FEM) or other 3D techniques. Furthermore, the techniques herein can model and analyze not only large blood vessels but also microvasculature, thus determining the state of occlusion in large blood vessels and the state of microvascular disease in small blood vessels.

[0041] More generally, the model 212 is configured to determine assimilated blood flow data over a sampling period for one or more vessels within the 3D or 1D vascular tree geometric model. Such determining may include determining pressure and / or flow velocity of multiple connected vessels in the 3D or 1D vascular tree geometric model.

[0042] In some examples, the lumped parameter boundary condition parameters are determined by the fluid dynamics machine learning model 212 for one or more blood vessels in the vascular examination region. In some examples, the fluid dynamics machine learning model 212 determines a lumped parameter model of blood flow for a first blood vessel, and determines a lumped parameter model of blood flow for each blood vessel branching off from the first blood vessel, either of which may be stored as assimilated blood flow data.

[0043] The assimilated blood flow data from the blood flow data generator 210 and the 3D vascular tree model (or 1D vascular tree model) are provided to a computational fluid dynamics model 214 that may apply physics-based processes to determine a state of vascular occlusion of one or more blood vessels in the 3D vascular tree model (or 1D vascular tree model) and / or a state of microvascular disease of one or more blood vessels. In some examples, the computational fluid dynamics model includes one or more of a multi-scale 3D Navier-Stokes simulation using a low-dimensional (lumped parameter) model, a low-dimensional Navier-Stokes (1D) simulation using a low-dimensional model that is a low-dimensional model derived from a graph theory framework that relies on 1D nonlinear theoretical models, or a low-dimensional model simulation (lumped parameter model, 0D) model of the entire segmented vascular tree model. In the illustrated example, the computational fluid dynamics model includes at least a 3D high fidelity training model 211 and a graph theory information low-dimensional model 213, by way of examples herein.

[0044] In some examples, the computational fluid dynamics model 214 is configured to determine an FFR, an iFR, and / or a QFR of one or more blood vessels in the 3D vascular tree model or the 1D vascular tree model from the blood flow data. In some examples, the computational fluid dynamics model 214 is configured to determine a state of vascular occlusion from the FFR, an iFR, and / or a QFR of the one or more blood vessels. In some examples, the computational fluid dynamics model 214 is configured to determine a coronary flow reserve (CFR) of the one or more blood vessels from the blood flow data from one or more physiological conditions (baseline and hyperemia) and to determine a state of microvascular disease from the CFR of the one or more blood vessels. Determining the state of vascular occlusion includes determining the presence of a stenosis in the one or more blood vessels. Determining the state of microvascular disease includes determining a lumped parameter model on a boundary of the blood vessel in the vascular examination region.

[0045] 3, a process 300 for assessing coronary artery disease as may be performed by the system 100 is shown. 2D angiographic image data is obtained by the medical imaging device 116 and provided to the computing device 102 in process 302, where optionally pre-processing operations may be performed. In a training mode, the 2D angiographic image data may be captured clinical angiographic image data, but such training data may further include synthetic image data, geometrically adjusted image data, etc., as shown in FIG.

[0046] In one example, training of the vessel segmentation machine learning model 206 was performed on 462 clinical angiographic images (see, e.g., FIG. 8 ) that were expanded via a combination of geometric transformations (zoom, horizontal flip, vertical flip, rotation, and / or shear) to form over 500,000 angiographic images and 461 synthetic angiographic images (see, e.g., FIG. 9 ) that were expanded via these geometric transformations to an additional set of over 500,000 images. The clinical angiographic images may include multiple frames of time series data with segmentation for extraction speed across the entire vascular tree, as shown. In some examples, the vessel segmentation machine learning model 206 includes a synthetic image generator configured to generate synthetic images using a combination of transformations such as flipping, shearing, rotation, and / or zoom. In any event, these numbers of images are provided only as empirical examples, and any suitable number of training images, captured or synthesized, may be used. In a training mode, image data is applied to the CAD assessment machine learning framework 110 for generation of two models, a vessel segmentation machine learning model 206 and a fluid dynamics machine learning model 212. In a diagnosis mode, image data is applied to the CAD assessment machine learning framework 110 for classification and diagnosis of CAD.

[0047] In process 304, the CAD assessment machine learning framework 110 applies the received image data to the vessel segmentation machine learning model, e.g., via the APN 202 and the semantic NN 207 of the vessel segmentation machine learning model 206, to generate a 2D segmented vessel image. The CAD assessment machine learning framework 110, such as via the 3D modeler 208, receives the 2D segmented vessel image and in process 306 generates a 3D segmented vessel tree model or a 1D segmented vessel tree model.

[0048] In process 308, the 3D segmented vascular tree model or the 1D segmented vascular tree model is applied to the fluid dynamics machine learning model 212 of the blood flow data generator 210, and assimilated blood flow data is generated over a sampling period of one or more vessels in the 3D vascular tree model or the 1D segmented vascular tree model.

[0049] In process 310, the assimilated blood flow data and the 3D vascular tree model or the 1D segmented vascular tree model are applied to a computational fluid dynamics model 214 that evaluates the data using either the 3D segmented vascular tree model or the 1D vascular tree model to determine vascular health, such as by determining the state of vascular occlusion via indices such as FFR, iFR, QFR, etc., by solving either the 3D Navier-Stokes equations or a reduced dimensional model based on graph theory. If data for two hemodynamic states (e.g., baseline and hyperemic states) is available, the state of microvascular disease, or CFR, is determined from lumped parameter values ​​at the boundary states for each hemodynamic state.

[0050] The process 400 shown in FIG. 4 is an exemplary implementation of the process 304 for generating a 2D segmented vascular image from angiographic image data that may be performed by the angiographic processing network 202 and the semantic NN 207 of the vascular segmentation machine learning model 206. Initially, in process 302, 2D angiographic image data of a target vascular region is received. In process 402, image preprocessing (including an image size normalization process, and / or a pixel intensity normalization process) is applied prior to further image processing via the angiographic processing network implemented with a convolutional neural network (e.g., APN 202), and further preprocessing may include a nonlinear filtering process, noise removal, and / or contrast enhancement, which, when combined with process 404, filters out objects such as catheters and bone structures. That is, in process 404, the preprocessed 2D angiographic image is applied to a second convolutional neural network (CNN) trained using clinical angiographic image data, synthetic image data, and / or geometrically modified image data (e.g., semantic NN 207). FIG. 5 illustrates an exemplary CNN framework 500 comprised of an angiogram processing network (APN) 503 in combination with a semantic NN in the form of a Deeplab v3+ network 507 configured to generate 2D segmented vascular images in process 408. The Deeplab v3+ network 507 is a semantic segmentation deep learning technique that includes an encoder 502 and a decoder 504. A series of angiogram images 501 are provided as input to the angiogram processing network 503, which consists of several 3×3 and 5×5 convolutional layers, which apply nonlinear filters that perform contrast enhancement, boundary sharpening, and other image processing functions. The APN 503, together with the semantic NN 507, form a vascular segmentation machine learning model, such as model 206. The APN 503, which feeds into the encoder 502, applies atlas convolutions to the image data, evolving a velocity that controls the effective field of view of each convolutional layer. The greater the velocity, the larger the area of ​​image capture for convolutions and low-level feature extraction as may be performed on each input image.For example, 6, 12, and 18 speeds may be used to affect different fields of view and capture different features, e.g., features at different resolutions. Using atlas or dilated convolutions, dilated samples of an image, or image portions, are convolved into a smaller image. The encoder 502 determines low-level features using several atlas convolution strides, and then applies 1×1 (depthwise separable) convolutions to combine the output of many different atlas convolutions. This creates a single matrix of features associated with different classes that is input to the decoder 504 along with the high-level features determined from the 1×1 convolutions applied to the original input image. The encoder 502 convolutions may be performed using atlas or depthwise convolutions in some examples. The decoder 504 concatenates the low-level and high-level features into a stacked matrix, and then applies a transposed convolution to this matrix using a fractional stride size to assign a class label to each pixel based on the spatial and feature information. These transposed convolutions generate a probability map that determines the likelihood that each pixel belongs to the background or vascular class. As shown, a softmax function is applied to generate an output segmented 2D image 505.

[0051] 6, a process 600 that may be performed by the 3D segmented vascular tree model generator 204 to generate a 3D segmented vascular tree model from a generated 2D segmented vascular image 602. As shown, any one of four different pipelines may be used to generate the 3D segmented vascular tree model, from which, by way of example, a 1D segmented vascular tree model may also be obtained.

[0052] In a first configuration, in process 603, the 3D modeler 208 receives the 2D segmented vascular images and finds the centerline of each of the 2D segmented vascular images. In process 604, the 3D modeler 208 uses a geometric tool, which may include epipolar geometry, projective geometry, Euclidean geometry, or any related geometric tool, to co-locate points from each of the 2D segmented vascular images and in process 606 triangulate the 3D points having projections (backprojections) that map onto the co-located points. A local radius of the vessel is determined from each 2D segmented vessel and these radius vectors are projected onto the 3D centerline. From there, in process 608, the 3D modeler 208 determines the vessel contour based on the triangulated 3D points and in process 610, a 3D vascular tree model is generated. From there, a 1D vascular tree model is generated in process 620.

[0053] In a second configuration, the 3D modeler 208 generates a number of 3D rotation matrices from the 2D segmented vascular images, in process 612. The 3D modeler 208 then generates a 3D segmented vascular tree model, in process 614, by solving a linear least-squares simultaneous equation that maps the number of 2D segmented vascular images into 3D space.

[0054] In a third configuration, the 3D modeler 208 forward projects voxels in 3D space onto the multiple 2D segmented vascular images, process 616, and identifies a set of 3D voxels that project within the multiple 2D segmented vascular images, process 618. The resulting binary volume is then smoothed to ensure realistic vascular boundaries.

[0055] In a fourth configuration, the active contour model is used to reconstruct the 3D geometry of the vessel. A finishing centerline of each 2D segmented vessel image is performed in process 622. The endpoints of each vessel are identified in the multiple 2D segmented vessel images and back projected in process 624 to identify the endpoints in 3D space. A 3D cylinder is drawn between these 3D endpoints, and in process 626, the external and internal forces of this cylinder are defined by the material properties, the imaging parameters of the system, and the reprojection error between the cylinder and the multiple 2D segmented images, as contour modeling, so that the projections are deformed until they match the vessel shape of each 2D image. In process 628, a reprojection deformation is performed to force the 2D image into a 3D cylinder, where deformations may be performed. The cylinder is deformed by the forces to minimize the reprojection error. Process 628 may be repeated for all branches of the vessel until the complete coronary artery tree is reconstructed.

[0056] The fluid dynamics machine learning model 212 herein, in some examples, is implemented as a neural network based on physics information, specifically, a neural network that can encode the underlying physical laws governing a given data set and can be described by partial differential equations. For example, the fluid dynamics machine learning model may include partial differential equations for the 3D Navier-Stokes equations, which are a set of four partial differential equations to balance mass and momentum, where the unknown fields are the 3D velocity vector (vx, vy, vz) and the scalar p.

[0057] In one example, a solution for blood flow within a 3D vascular tree model is generated by Equations 1 and 2, using the incompressible 3D Navier-Stokes equations of blood flow.

number

number

[0058] Alternatively, in some examples, the fluid dynamics machine learning model 212 may include a graph theory based reduced dimensional model, as shown in FIG. 7. The graph theory based reduced dimensional model may be executed in a machine learning mode or an analytical mode. In the machine learning mode, training data 701 on measurements such as blood flow, stenosis geometry, FFR, iFR, QFR, etc. are used to define a dense graph of discrepancies 702 between ground truth data and a low fidelity model of blood flow, given by a 1D nonlinear theoretical model 703. The ground truth data can be provided either by in silico simulation of 3D Navier-Stokes equations 704 or by in vivo data acquired in a catheterization lab 705. The dense graph of discrepancies 702 may be generated by analyzing the displacement of diameters of multiple stenoses, lengths, eccentricities, blood flow through the stenoses, or combinations thereof. Once the dense graph of discrepancies is generated, a reduced dimensional model 706 may be derived via nonlocal calculus, deep neural networks, or direct traversal of the graph vertices. The low-dimensional model may be an algebraic equation or an ordinary or partial differential equation. In the analysis mode, the input 707 to the low-dimensional model is the geometry given by the 1D segmented vascular tree extracted by process 610 and the boundary conditions of blood flow and pressure extracted via the data assimilation process 308. The low-dimensional model 706 and the input 707 create the desired state of vascular occlusion 708 (anatomical and functional with FFR calculation, iFR calculation, or QFR calculation) and the state of microvascular disease. The 1D nonlinear theoretical model 703 is a 1D nonlinear Navier-Stokes model, which includes two partial differential equations for mass balance and momentum balance, with the unknowns being blood flow Q and cross-sectional mean pressure P.

[0059] In one example, a blood flow solution obtained with a 1D nonlinear theoretical model is generated using the mass and momentum interactions of an incompressible Newtonian fluid by the following simultaneous equations, Equations 3 and 4:

number

[0060] In any case, the fluid dynamics machine learning models herein may be formed with data-driven algorithms to infer solutions to these general nonlinear partial differential equations via surrogate classification models based on physical information. Information about the principle physical laws governing the time-dependent dynamics of the system, or about empirically verified rules or other domain expertise, may be used as a regularization agent to restrict the space of permissible solutions to a manageable size. In return, encoding such structured information into the machine learning model amplifies the information content of the data the algorithm sees, allowing it to steer quickly towards a good solution and generalize even when there are only a few training examples. Furthermore, the reduced-dimensional models proposed herein are trained using graph theory and the mismatch between a low-fidelity 1D nonlinear model of blood flow and ground truth data provided by either a 3D high-resolution Navier-Stokes model or in vivo anatomical and hemodynamic data to accurately and efficiently capture the hemodynamics around the stenosis. In various examples, the reduced-dimensional model is defined from the mismatched graph via one of three methods: a) CNN, b) nonlocal calculus, or c) graph exploration using a traversal algorithm (see, e.g., Banerjee et al., A graph theoretic framework for representation, exploration and analysis on computed states of physical systems, Computer Methods in Applied Mechanics and Engineer, 2019, incorporated herein by reference).

[0061] In one example, the fluid mechanics machine learning model is configured to include Hidden Fluid Mechanics (HFM), a deep learning framework based on physics information that can encrypt a class of physical laws governing fluid motion, namely the Navier-Stokes equations, as described in A Navier-Stokes Informed Deep Learning Framework for Assimilating Flow Visualization Data, Raissi et al., Hidden Fluid Mechanics, dated August 13, 2018, which is incorporated herein by reference. In one example, the fluid mechanics machine learning model applies underlying conservation laws (i.e., mass, momentum, energy) to infer hidden quantities of interest, such as velocity and pressure fields, from 3D vascular tree models generated at different times from angiographic image data taken at different times. The fluid mechanics machine learning model may apply algorithms that are independent of geometry or initial and boundary conditions. This allows the HFM configuration to be very flexible in selecting the type of vascular image data that can be used for training and diagnosis by the model. The fluid mechanics machine learning model is trained to predict pressure and velocity values ​​of blood flow in both two and three dimensions of the imaged blood vessels. Such information can be used to determine other physically related properties, such as pressure or wall shear stress within the artery.

[0062] In some examples, the computational fluid dynamics model is configured to determine a lumped parameter model attached to each vessel of the 3D vascular tree model or the 1D vascular tree model. The computational fluid dynamics model may include a set of lumped parameter models (LPMs) for different vessels, as shown in FIG. 10A. An exemplary LPM 1000 is provided for a heart model coupled to the inflow surface of a 3D vascular tree 1002. The vascular tree 1002 is formed of a number of different vessels labeled with different letters, A inlet, B-H aortic outlet, and a-k coronary outlet. An LPM 1006 is used for each outlet B-H representing the microcirculation of the vessels other than the coronary arteries. An LPM 1008 is provided for the coronary outlets a-k and coupled to the LPM 1000 representing the heart. The parameters of this model are estimated from the patient's data on blood flow and pressure (measured in the catheterization lab or estimated using data assimilation techniques described in 308 or using morphometric considerations such as Murray's Law). The analysis may be performed in a closed loop configuration using the LPM 1004, which includes both the left and right sides of the heart. Models of this nature may be used to calculate hemodynamics under pulsatile conditions. A second example of an LPM is shown in FIG. 10B. Here, the inflow boundary conditions may be defined by either the patient's mean arterial pressure with average blood flow measured or estimated from dynamic angiography data. The patient-specific outflow boundary conditions may be defined by the blood flow down each vessel estimated via a fluid dynamics machine learning process 308 that assimilates blood flow data over a sampling period, or by the LPM (resistance) coupled to each of the outlet faces of the coronary artery tree. Knowing the calculated solutions for pressure and flow in the vascular tree, the LPM for each tree (3D or 1D) of the vascular tree may also be estimated, assuming that the pressure gradient across the LPM drops to a constant level of capillary pressure. Models of this nature may be used to simulate hemodynamics under steady-state conditions. This process may be repeated for each available hemodynamic state (e.g., baseline and hyperemia). LPMs are used to represent cardiovascular regions where full detail of the blood flow solution is not required, but it is important that the model includes the relationships between pressure, blood flow, and possibly volume in these regions.They are ideal in regions where detailed spatial information about the vascular geometry is neither available nor fundamentally important. Since their parameters have a fundamental effect on the pressure and velocity fields of the vascular tree model, the parameters must be adjusted to achieve physiological values ​​that are consistent with the data assimilated from the patient's images and additional clinical records available for the individual in question. When insufficient data is available for parameterization of a particular patient, data from the scientific literature on expected values ​​can be used to help determine appropriate parameters.

[0063] In one example, the anatomical and functional assessment of CAD follows a workflow 1100 shown in FIG. 11. Several angiographic images 1101 taken at different orientations are fed to a machine learning segmentation module 1102 (corresponding to processes 302, 304, and 402 and 404), which automatically creates 2D segmented images of the angiographic images 1103. These images are then fed to an algorithm 1104 that generates 3D and 1D segmented vascular tufts 1105 by a combination of the processes given in FIG. 6. This workflow is used to automatically characterize the lumen diameter and therefore the anatomical severity of CAD. Furthermore, a series of angiographic images 1106 defining the transport of dye down regions of interest in the vascular tree are fed to a series of fluid dynamics machine learning algorithms 1107 that assimilate information about the blood flow velocity and / or pressure of each vessel of interest in the vascular tree. The system generates velocity and pressure boundary conditions 1108 which, along with the vascular tree 1105, are fed as input to a graph theory derived reduced dimensional model 1109, ultimately producing the desired functional metrics of the CAD 1110 including FFR, iFR, QFR and microvascular resistance.

[0064] Additional Aspects Aspect 1. A computer-implemented method for assessing coronary artery disease, comprising: (a) receiving, by one or more processors, angiographic image data of a vascular examination area of ​​a subject, the angiographic image data including angiographic images captured over a sampling period; (b) applying, by the one or more processors, the angiographic image data to a vascular segmentation machine learning model and generating, using the vascular segmentation machine learning model, a two-dimensional (2D) segmented vascular image of the vascular examination area; (c) generating, by the one or more processors, a three-dimensional (3D) segmented vascular tree geometric model of blood vessels in the vascular examination region from the 2D segmented vascular images; (d) applying, by the one or more processors, the 3D segmented vascular tree geometric model to a fluid dynamics machine learning model to assimilate blood flow data over a sampling period of one or more blood vessels in the vascular examination region using the fluid dynamics machine learning model; (e) applying, by the one or more processors, the 3D segmented vascular tree geometric model and the assimilated blood flow data to a 3D high-fidelity computational fluid dynamics model; (f) determining, by the one or more processors, a vascular occlusion status of one or more blood vessels within the vascular examination region.

[0065] Aspect 2. The computer-implemented method of aspect 1, further comprising determining, by one or more processors, a microvascular disease status of one or more blood vessels within the vascular examination region by performing (a)-(e) under at least two different hemodynamic conditions.

[0066] Aspect 3. The computer-implemented method of aspect 1, wherein the vascular segmentation machine learning model is a convolutional neural network.

[0067] Aspect 4. The computer-implemented method of aspect 1, further comprising applying, by the one or more processors, at least one of a noise removal process, a linear filtering process, an image size normalization process, and a pixel intensity normalization process to the received angiographic image data to create filtered angiographic image data.

[0068] Aspect 5. The computer-implemented method of embodiment 4 further includes providing the filtered angiographic image data to an angiographic processing network (APN) trained to remove low contrast images, catheters, and / or overlapping Bony structures, which address the main challenges of angiographic image data.

[0069] Aspect 6. The computer-implemented method of embodiment 5, further comprising feeding the output of the APN to a semantic image segmentation to create an automatic binary 2D segmented vascular image.

[0070] Aspect 7. Generating a 3D segmented vascular tree geometric model includes: Finding the centerline of each of the 2D segmented vessel images; co-locating points from each of the 2D segmented vascular images using epipolar geometry; triangulating the 3D points having projections that map onto the co-located points; and determining a contour of the vessel based on the triangulated 3D points.

[0071] Aspect 8. Generating a 3D segmented vascular tree geometric model includes: generating a plurality of 3D rotation matrices from the plurality of 2D segmented vascular images; 2. The computer-implemented method of embodiment 1, comprising: generating a 3D segmented vascular tree geometric model by solving a linear least-squares simultaneous equation that maps a plurality of 2D segmented vascular images into a 3D space.

[0072] Aspect 9. Generating a 3D segmented vascular tree geometric model includes: forward projecting voxels in 3D space onto a plurality of 2D segmented vascular images; 2. The computer-implemented method of embodiment 1, comprising: identifying a set of 3D voxels that project within the plurality of 2D segmented vascular images.

[0073] Aspect 10. Generating a 3D segmented vascular tree geometric model includes: 2. The computer-implemented method of embodiment 1, comprising using active contours to deform a cylindrical geometry via internal and external forces into a plurality of segmented 2D images.

[0074] Aspect 11. The computer-implemented method of embodiment 1, further comprising applying, by the one or more processors, at least one of a smoothing algorithm and a surface spline approximation algorithm to the 3D segmented vascular tree geometric model.

[0075] Aspect 12. The computer-implemented method of aspect 1, generating a 3D segmented vascular tree geometric model by performing backprojection of 2D segmented vascular images.

[0076] Aspect 13. The computer-implemented method of aspect 1, wherein the fluid dynamics machine learning model includes at least one of the following types of networks: a convolutional neural network (CNN), an autoencoder, or a long short-term memory (LSTM).

[0077] Aspect 14. The computer-implemented method of aspect 1, wherein the fluid dynamics machine learning model is a Navier-Stokes informed deep learning framework configured to determine pressure and velocity data over 3D vascular space.

[0078] Aspect 15. A computer-implemented method as described in aspect 14, wherein the deep learning framework based on Navier-Stokes information includes one or more of the following types of methods: Kalman filtering, neural networks based on physical information, iterative assimilation algorithms based on contrast arrival times to anatomical landmarks, and TIMI frame counting.

[0079] Aspect 16. The computer-implemented method of aspect 1, wherein determining blood flow data over the sampling period includes determining pressure and flow velocity data for one or more blood vessels over the sampling period.

[0080] Aspect 17. The computer-implemented method of aspect 1, wherein determining blood flow data over a sampling period includes determining pressure and flow velocity data for a plurality of connected blood vessels in the vascular examination area.

[0081] Aspect 18. The computational fluid dynamics model comprises: The computer-implemented method of aspect 1 includes one or more of a multi-scale 3D Navier-Stokes simulation using a low-dimensional (lumped parameter) model, a low-dimensional Navier-Stokes (1D) simulation using a low-dimensional model, or a low-dimensional model simulation (lumped parameter model, 0D) model of the segmented vascular tree geometric model.

[0082] Aspect 19. The computer-implemented method of aspect 1, wherein the lumped parameter boundary condition parameters are determined by a fluid dynamics machine learning model of one or more blood vessels in the vascular examination region.

[0083] Aspect 20. The computer-implemented method of aspect 19, further comprising: determining a lumped parameter model of blood flow in the first vessel; and determining a lumped parameter model of blood flow in each vessel branching from the first vessel.

[0084] Aspect 21. determining a fractional flow reserve (FFR), an instantaneous fractional flow reserve (iFR), or a quantitative flow ratio (QFR) for one or more vessels from the flow data; 2. The computer-implemented method of any of aspects 1, further comprising determining a state of vascular occlusion from an FFR, iFR, or QFR of one or more blood vessels.

[0085] Aspect 22. determining a coronary flow reserve (CFR) for one or more vessels from blood flow data from one or more physiological conditions; 2. The computer-implemented method of embodiment 1, further comprising determining a microvascular disease status from the CFR of one or more blood vessels.

[0086] Embodiment 23. The computer-implemented method of embodiment 22, wherein the one or more physiological conditions include a baseline physiological condition and a hyperemic physiological condition.

[0087] Aspect 24. The computer-implemented method of aspect 1, wherein determining the state of vascular occlusion includes determining the presence of a stenosis in one or more blood vessels.

[0088] Aspect 25. The computer-implemented method of aspect 1, wherein determining the state of microvascular disease includes determining a lumped parameter model on a boundary of a blood vessel in the vascular examination region at a plurality of hemodynamic conditions.

[0089] Aspect 26. The computer-implemented method of aspect 1, further comprising supplying a plurality of synthetic angiography images, a plurality of clinical angiography images, and a plurality of augmented angiography images to train the vascular segmentation machine learning model.

[0090] Aspect 27. The computer-implemented method of aspect 1, wherein the angiographic image data includes angiographic images captured over a sampling period, including angiographic images captured during a baseline condition and angiographic images captured during a pharmacologically-induced hyperemic condition.

[0091] Aspect 28. A computer-implemented method for assessing coronary artery disease, comprising: (a) receiving, by one or more processors, a plurality of angiographic images of a vascular examination region of a subject, the angiographic images being captured over a sampling period, the vascular examination region including one or more blood vessels; (b) applying, by one or more processors, the angiographic image to a vessel segmentation machine learning model and generating, using the vessel segmentation machine learning model, a two-dimensional (2D) segmented vessel image of one or more vessels; (c) generating, by the one or more processors, a one-dimensional (1D) segmented vascular tree geometric model of one or more blood vessels from the 2D segmented vascular image; (d) applying, by the one or more processors, the 1D segmented vascular tree geometric model to a fluid dynamics machine learning model and assimilating the blood flow data over a sampling period of one or more blood vessels using the fluid dynamics machine learning model; (e) applying, by the one or more processors, the 1D segmented vascular tree geometric model and the assimilated blood flow data to a graph theory based reduced dimensional model based on a computational fluid dynamics model; (f) determining, by the one or more processors, a status of the vascular occlusion.

[0092] Aspect 29. The computer-implemented method of aspect 28, further comprising determining, by one or more processors, a microvascular disease status of one or more blood vessels within the vascular examination area by performing (a)-(e) under at least two different hemodynamic conditions.

[0093] Aspect 30. The computer-implemented method of aspect 28, wherein the angiographic images captured over the sampling period include angiographic images captured during a baseline state and angiographic images captured during a hyperemic state.

[0094] A computer-implemented method as described in Aspect 30, in which generating a 1D segmented vascular tree geometric model of a blood vessel in a vascular inspection area from a 2D segmented vascular image includes creating a skeletonization of the 3D segmented vascular tree geometric model given the 3D space of the blood vessels contained in the 3D segmented vascular tree geometric model and the paths / centerlines in a series of 2D cross-sectional contours separated by any distance along each path / centerline of the tree.

[0095] Aspect 32. The computer-implemented method of aspect 28, wherein the vascular segmentation machine learning model is a convolutional neural network.

[0096] Aspect 33. The computer-implemented method of embodiment 28, further comprising applying, by the one or more processors, a noise removal process, a linear filtering process, an image size normalization process, and a pixel intensity normalization process to the received angiography image.

[0097] Aspect 34. The computer-implemented method of aspect 28, wherein the fluid dynamics machine learning model includes at least one type of network of the type of convolutional neural network (CNN), autoencoder, or long short-term memory (LSTM), or a low-dimensional model of blood flow and pressure based on graph theory.

[0098] Aspect 35. The computer-implemented method of aspect 28, wherein the fluid dynamics machine learning model is a deep learning framework based on Navier-Stokes information configured to determine pressure and velocity data over 1D vessel space.

[0099] Aspect 36. A computer-implemented method as described in aspect 35, wherein the deep learning framework based on Navier-Stokes information includes one or more of the following types of methods: Kalman filtering, neural networks based on physical information, iterative assimilation algorithms based on contrast arrival times to anatomical landmarks, and TIMI frame counting.

[0100] Aspect 37. The computer-implemented method of aspect 28, wherein the fluid dynamics machine learning model is a low-dimensional model based on graph theory obtained by defining a dense graph of discrepancies between ground truth data and a low-fidelity model of blood flow given a 1D nonlinear theoretical model.

[0101] Aspect 38. The computer-implemented model of aspect 37, wherein the ground truth data used to define the dense graph can be either an in silico simulation of 3D Navier-Stokes equations of in vivo data acquired in a catheterization laboratory.

[0102] Aspect 39. The computer-implemented model of aspect 37, wherein the vertices of the dense graph are defined by investigating discrepancies between ground truth data and a low-fidelity 1D nonlinear model of blood flow, including displacement of stenosis diameter, length, eccentricity, blood flow through the stenosis, and combinations thereof.

[0103] Aspect 40. The computer-implemented model according to aspect 37, wherein the graph-theoretic based reduced-dimensional model is obtained by non-local calculus, deep neural networks, or direct traversal of the vertices of the generated graph.

[0104] Aspect 41. The computer-implemented model of aspect 37, wherein the graph-theoretically based reduced-dimensional model may be an algebraic equation or an ordinary or partial differential equation.

[0105] Aspect 41. The computer-implemented method of aspect 28, wherein determining blood flow data over the sampling period includes determining pressure and flow velocity data for one or more blood vessels over the sampling period.

[0106] Embodiment 43. The computer-implemented method of embodiment 42, wherein the one or more blood vessels include a plurality of connected blood vessels.

[0107] Aspect 44. The computer-implemented method of aspect 28, wherein the computational fluid dynamics model includes a reduced-dimensional model simulation based on graph theory, where the input is provided by a 1D segmented vascular tree geometric model and the boundary conditions of blood flow and pressure are assimilated by a computational fluid dynamics machine learning model.

[0108] Embodiment 45. The computer-implemented method of embodiment 28, further comprising determining a lumped parameter model of blood flow in each of the one or more blood vessels.

[0109] Aspect 46. determining a fractional flow reserve (FFR), an instantaneous fractional flow reserve (iFR), or a quantitative flow ratio (QFR) for one or more vessels from the flow data; 2. The computer-implemented method of any of aspects 1, further comprising determining a state of vascular occlusion from an FFR, iFR, or QFR of one or more blood vessels.

[0110] Aspect 47. determining a coronary flow reserve (CFR) for one or more vessels from blood flow data from one or more physiological conditions; 2. The computer-implemented method of embodiment 1, further comprising determining a microvascular disease status from the CFR of one or more blood vessels.

[0111] Aspect 48. The computer-implemented method of aspect 1, wherein determining the state of vascular occlusion comprises determining the presence of stenosis in 28 or more blood vessels.

[0112] Aspect 49. The computer-implemented method of aspect 28, further comprising supplying a plurality of synthetic angiography images, a plurality of clinical angiography images, and a plurality of augmented angiography images to train the vascular segmentation machine learning model.

[0113] Aspect 50. A method of implementing a method for detecting a vascular defect in a vascular examination region of a subject, comprising: receiving, by the one or more processors, angiographic image data of a vascular examination region of a subject, the angiographic image data including angiographic images captured over a sampling period; applying, by the one or more processors, the angiographic image data to a vascular segmentation machine learning model and generating, using the vascular segmentation machine learning model, a two-dimensional (2D) segmented vascular image of the vascular examination region; and generating, by the one or more processors, a three-dimensional (3D) segmented vascular tree geometric model of blood vessels in the vascular examination region from the 2D segmented vascular image. and one or more computer-readable memories storing instructions that cause the one or more processors to: generate a 3D segmented vascular tree geometric model based on a vascular occlusion model; apply, by the one or more processors, the 3D segmented vascular tree geometric model to a fluid dynamics machine learning model to assimilate blood flow data over a sampling period of one or more blood vessels in the vascular examination region using the fluid dynamics machine learning model; apply, by the one or more processors, the 3D segmented vascular tree geometric model and the assimilated blood flow data to a 3D high-fidelity computational fluid dynamics model; and determine, by the one or more processors, a state of vascular occlusion of the one or more blood vessels in the vascular examination region.

[0114] Aspect 50. A method according to claim 1, further comprising: receiving, by the one or more processors, a plurality of angiographic image data of a vascular examination region of a subject, the angiographic images being captured over a sampling period, the vascular examination region including one or more blood vessels; applying, by the one or more processors, the angiographic images to a vascular segmentation machine learning model and generating, using the vascular segmentation machine learning model, two-dimensional (2D) segmented vascular images of the one or more blood vessels; and generating, by the one or more processors, one-dimensional (1D) segmented vascular images of the one or more blood vessels from the 2D segmented vascular images. and one or more computer-readable memories storing instructions that cause the one or more processors to: generate a 1D segmented vascular tree geometric model; apply, by the one or more processors, the 1D segmented vascular tree model to a fluid dynamics machine learning model to assimilate blood flow data over a sampling period of one or more vessels using the fluid dynamics machine learning model; apply, by the one or more processors, the 1D segmented vascular tree model and the assimilated blood flow data to a graph theory based reduced dimensional model that is based on a computational fluid dynamics model; and determine, by the one or more processors, a state of vascular occlusion.

[0115] Throughout this specification, multiple instances may implement components, operations, or structures described as a single instance. Although individual operations of one or more methods have been illustrated and described as separate operations, one or more of the individual operations may be performed simultaneously, and the operations need not be performed in the order illustrated. Structures and functions presented as separate components in an example configuration may be implemented as a combined structure or component. Similarly, structures and functions presented as a single component may be implemented as separate components. These and other variations, modifications, additions, and improvements are within the scope of the subject matter of this specification.

[0116] Further, certain embodiments are described herein as including logic or a number of routines, subroutines, applications, or instructions. These may constitute either software (e.g., code embodied on a non-transitory machine-readable medium) or hardware. In hardware, routines, etc. are tangible units capable of performing certain operations and may be configured or arranged in a particular manner. In an exemplary embodiment, one or more computer systems (e.g., standalone, client, or server computer systems), or one or more hardware modules of a computer system (e.g., a processor or processors), may be configured as hardware modules that operate by software (e.g., an application or portion of an application, such as the contrast injection system shown in FIG. 2) to perform certain operations described herein.

[0117] In various embodiments, a hardware module can be implemented mechanically or electronically. For example, a hardware module can include dedicated circuitry or logic that is permanently configured to perform certain operations (e.g., a special-purpose processor such as a field programmable gate array (FPGA) or an application specific integrated circuit (ASIC)). A hardware module can also include programmable logic or circuitry that is temporarily configured by software to perform certain operations (e.g., implemented in a general-purpose processor or other programmable processor). It will be appreciated that the decision as to whether a hardware module is implemented mechanically, with dedicated and permanently configured circuitry, or with temporarily configured circuitry (e.g., configured by software) can be determined based on cost and time considerations.

[0118] Thus, the term "hardware module" should be understood to encompass tangible entities, that is physically constructed, permanently configured (e.g., hardwired), or temporarily configured (e.g., programmed) to operate in a certain manner or to perform certain operations described herein. Considering embodiments in which the hardware modules are temporarily configured (e.g., programmed), each of the hardware modules need not be configured or instantiated at any one instance of time. For example, if the hardware modules include a general-purpose processor configured using software, the general-purpose processor may be configured as different hardware modules at different times. Thus, the software may configure the processor, for example, to configure a particular hardware module at one time and another hardware module at another time.

[0119] Hardware modules can provide information to and receive information from other hardware modules. Thus, the hardware modules described herein can be understood as communicatively coupled. When such hardware modules are present simultaneously, communication can be achieved via signal transmission (e.g., through appropriate circuits and buses) connecting the hardware modules. In embodiments in which multiple hardware modules are configured or instantiated at different times, communication between such hardware modules can be achieved, such as through storage and retrieval of information in a memory structure to which the multiple hardware modules have access. For example, a hardware module can perform an operation and store the output of that operation in a memory device to which the hardware module is communicatively coupled. An additional hardware module can then later access the memory device to retrieve and process the stored output. Hardware modules can also initiate communication with input or output devices to operate on resources (e.g., collect information).

[0120] Various operations of the example methods described herein may be performed, at least in part, by one or more processors that are temporarily (e.g., by software) or permanently configured to perform the associated operations. Such processors, whether temporarily or permanently configured, may constitute processor-implemented modules that operate to perform one or more operations or functions. Modules referred to herein may, in some example embodiments, include processor-implemented modules.

[0121] Similarly, methods or routines described herein may be at least partially processor-implemented. For example, at least some of the operations of a method may be performed by one or more processors or processor-implemented hardware modules. Certain performance of an operation may reside not only in a single machine, but also be distributed among one or more processors deployed across several machines. In some example embodiments, one or more processors may reside in a single location (e.g., in a home environment, in a work environment, or as a server farm), while in other embodiments, the processors may be distributed across multiple locations.

[0122] The reliable performance of an operation can be distributed among one or more processors that are not only present in a single machine, but also deployed across several machines. In some exemplary embodiments, one or more processors or processor-implemented modules may be present in a single location (e.g., in a home environment, in a work environment, or in a server farm). In other exemplary embodiments, one or more processors or processor-implemented modules may be distributed across multiple locations.

[0123] Unless otherwise indicated, descriptions herein using words such as "processing," "computing," "calculating," "determining," "presenting," and "displaying" may refer to machine (e.g., computer) operations or processes that manipulate or transform data represented as physical (e.g., electronic, magnetic, or optical) quantities in one or more memories (e.g., volatile memory, nonvolatile memory, or a combination thereof), registers, or other mechanical components that receive, store, transmit, or display information.

[0124] As used herein, any reference to "one embodiment" or "an embodiment" means that a particular element, feature, structure or characteristic described in connection with an embodiment is included in at least one embodiment. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment.

[0125] Some embodiments may be described using the terms "coupled" and "connected," along with their derivatives. For example, some embodiments may be described using the term "coupled" to indicate that two or more elements are in direct physical or electrical contact. However, the term "coupled" can also mean that two or more elements are not in direct contact with each other, but still co-operate or interact with each other. The embodiments are not limited in this context.

[0126] Those skilled in the art will recognize that numerous modifications, changes and combinations can be made to the above-described embodiments without departing from the scope of the present invention, and such modifications, changes and combinations should be considered within the scope of the inventive concept.

[0127] Thus, while the invention has been described with reference to specific examples, it will be apparent to those skilled in the art that these examples are illustrative only and are not intended to be limitations of the invention, and that modifications, additions, or deletions may be made to the disclosed embodiments without departing from the spirit and scope of the invention.

[0128] The foregoing description is set forth merely for clarity of understanding, and no unnecessary limitations should be understood therefrom, since modifications within the scope of the invention may be apparent to those skilled in the art.

Claims

1. 1. A computer-implemented method for assessing coronary artery disease, comprising: (a) receiving, by one or more processors, a plurality of angiographic images of a vascular examination region of a subject, the angiographic images being captured over a sampling period, the vascular examination region including one or more blood vessels; (b) applying, by the one or more processors, the angiographic image to a vessel segmentation machine learning model and generating, using the vessel segmentation machine learning model, a two-dimensional (2D) segmented vessel image of the one or more vessels; (c) generating, by the one or more processors, a one-dimensional (1D) segmented vascular tree geometric model of the one or more blood vessels or a three-dimensional (3D) segmented vascular tree geometric model of the one or more blood vessels from the 2D segmented vascular image; (d) applying, by the one or more processors, the 1D segmented vascular tree geometric model or a three-dimensional (3D) segmented vascular tree geometric model to a fluid dynamics machine learning model and assimilating blood flow data over a sampling period of the one or more blood vessels using the fluid dynamics machine learning model; and (e) applying, by the one or more processors, the 1D segmented vascular tree geometric model or the three-dimensional (3D) segmented vascular tree geometric model and the assimilated blood flow data to a graph theory-based reduced dimensional model based on a computational fluid dynamics model, and determining a state of vascular occlusion.

2. 2. The computer-implemented method of claim 1, further comprising: determining, by the one or more processors, a microvascular disease status of the one or more blood vessels in the vascular examination region by performing (a)-(e) under at least two different hemodynamic conditions.

3. 2. The computer-implemented method of claim 1, wherein the angiographic images captured over the sampling period include angiographic images captured during a baseline condition and angiographic images captured during a hyperemic condition.

4. 4. The computer-implemented method of claim 3, wherein generating the 1D segmented vascular tree geometric model of the blood vessels in the vascular examination area from the 2D segmented vascular image comprises creating a skeletonization of the 3D segmented vascular tree geometric model given the 3D space of the blood vessels contained in the 3D segmented vascular tree geometric model and paths / centerlines in a series of 2D cross-sectional contours separated by any distance along each path / centerline of the 3D segmented vascular tree geometric model.

5. 2. The computer-implemented method of claim 1, wherein the vessel segmentation machine learning model is a convolutional neural network.

6. 10. The computer-implemented method of claim 1, further comprising applying, by the one or more processors, a noise removal process, a linear filtering process, an image size normalization process, and a pixel intensity normalization process to the received angiographic image.

7. The fluid dynamics machine learning model includes at least one type of network: a convolutional neural network (CNN), an autoencoder, or a long short-term memory (LSTM), or a type of low-dimensional model of blood flow and pressure based on graph theory; the fluid dynamics machine learning model is a Navier-Stokes informed deep learning framework configured to determine pressure and velocity data over 1D vessel space; the Navier-Stokes information based deep learning framework comprises one or more of the following types of methods: Kalman filtering, neural networks based on physical information, iterative assimilation algorithms based on contrast arrival time to anatomical landmarks, and TIMI frame counting; or 2. The computer-implemented method of claim 1, wherein the fluid dynamics machine learning model is a graph-theoretic based low-dimensional model obtained by defining a dense graph of discrepancies between a 1D nonlinear theoretical model and ground truth data.

8. 8. The computer-implemented method of claim 7, wherein the ground truth data used to define the dense graph is obtained either by in-vivo data acquired in a catheterization lab or by in silico simulation of 3D Navier-Stokes equations.

9. 8. The computer-implemented method of claim 7, wherein the vertices of the dense graph are defined by examining discrepancies between the ground truth data and the 1D nonlinear model of blood flow, including displacement of stenosis diameter, length, eccentricity, blood flow through a stenosis, and combinations thereof.

10. the graph-theoretic low-dimensional model is a deep neural network; or The computer-implemented method of claim 7 , wherein the graph-theoretic based reduced-dimensional model comprises ordinary or partial differential equations.

11. The computer-implemented method of claim 1 , wherein determining blood flow data over the sampling period comprises determining pressure and flow velocity data for the one or more blood vessels over the sampling period.

12. 2. The computer-implemented method of claim 1, wherein the computational fluid dynamics model comprises a graph theory based reduced order model simulation and inputs include boundary conditions for the 1D segmented vascular tree geometric model blood flow and pressure.

13. determining a fractional flow reserve (FFR), an instantaneous fractional flow reserve (iFR), or a quantitative flow ratio (QFR) for the one or more vessels from the flow data; determining a status of the vascular occlusion from the FFR, iFR, or QFR of the one or more blood vessels; or determining a coronary flow reserve (CFR) of the one or more vessels from blood flow data from one or more physiological conditions; determining a microvascular disease status from the CFR of the one or more blood vessels; The computer-implemented method of claim 1 , further comprising:

14. The computer-implemented method of claim 1 , wherein determining the state of the vascular occlusion comprises determining the presence of a stenosis in the one or more blood vessels.

15. 10. The computer-implemented method of claim 1, further comprising providing a plurality of synthetic angiogram images, a plurality of clinical angiogram images, and a plurality of augmented angiogram images to train the vessel segmentation machine learning model.

Images