Tissue status graphic display system

Abstract

A system is provided for augmenting a three-dimensional (3D) model of the heart to indicate tissue status. The system accesses the 3D model of the heart, accesses two-dimensional (2D) images of slices of the tissue status of the heart, and accesses arrhythmia origin information. The system augments the 3D model with a representation of the origin based on the origin information. For each of a plurality of tissue status slices of the heart, the system augments a slice of the 3D model corresponding to the tissue status slice with a representation of the tissue status of the heart represented by the tissue status information for that tissue status slice. The system then displays a representation of the 3D model showing the origin of the arrhythmia and the tissue status of the heart.

Description

[Technical field]

[0001] (CROSS REFERENCE TO RELATED APPLICATIONS) This application claims the benefit of priority to U.S. Patent Application No. 63 / 231,022, filed August 9, 2021, which is incorporated by reference in its entirety. [Background technology]

[0002] Many cardiac diseases can cause symptoms, morbidity (e.g., syncope or stroke), and mortality. Common cardiac diseases caused by arrhythmias include inappropriate sinus tachycardia (IST), ectopic atrial rhythm, junctional rhythm, ventricular escape rhythm, atrial fibrillation (AF), ventricular fibrillation (VF), focal atrial tachycardia (focal AT), atrial micro-reentry, ventricular tachycardia (VT), atrial flutter (AFL), premature ventricular contractions (PVC), premature atrial contractions (PAC), atrioventricular nodal reentrant tachycardia (AVNRT), atrioventricular reentrant tachycardia (AVRT), permanent junctional reciprocating tachycardia (PJRT), and junctional tachycardia (JT). Sources of arrhythmias can include electrical rotors (e.g., VF), recurrent electrical focal sources (e.g., AT), anatomical-based reentry (e.g., VT), and the like. These sources are important factors in persistent or clinically significant manifestations. Arrhythmias can be treated by targeting the source of cardiac disease through ablation using a variety of techniques including radiofrequency energy ablation, cryoablation, ultrasound ablation, laser ablation, external radiation sources, directed gene therapy, etc. Because the source and site of cardiac disease varies from patient to patient, targeted treatment of even common cardiac diseases requires identifying the source of arrhythmia.

[0003] Unfortunately, conventional methods for reliably identifying the source and site of cardiac disease can be complex, cumbersome, and expensive. For example, one method uses an electrophysiology catheter with a multi-electrode basket catheter that is intravascularly inserted into the heart (e.g., the left ventricle) to collect measurements of the heart's electrical activity, such as during an induced VF episode. The measurements can then be analyzed to help identify the site of origin. Currently, electrophysiology catheters are expensive (and generally limited to single use) and can cause serious complications, including cardiac perforation and tamponade. Another method uses an external body surface vest with electrodes to collect measurements from the patient's body surface that can be analyzed to help identify the site of arrhythmia origin. Such body surface vests are expensive, complex and difficult to manufacture, and can interfere with the placement of defibrillator pads, which are required after inducing VF to collect measurements during the arrhythmia. Furthermore, analysis of the vest requires a computed tomography (CT) scan, and the body surface vest cannot sense the interventricular and interatrial septum, where approximately 20% of arrhythmia sources may originate.

[0004] Knowing the condition of cardiac tissue taken from a patient is useful in assessing the patient's cardiac function. Various approaches have been used to classify the condition of cardiac tissue as normal or abnormal. Abnormal cardiac tissue may indicate border zone tissue or scar tissue. Several techniques support determining the condition of cardiac tissue based on perfusion through the cardiac tissue, heart wall motion, electrical activity, and others. Although the condition of cardiac tissue can be used to identify regions of the heart whose function is normal or abnormal, the condition of cardiac tissue by itself cannot usually be used to determine the exact site of origin of an arrhythmia. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] U.S. Patent No. 10,860,754 [Patent Document 2] U.S. Pat. No. 10,952,794 [Patent Document 3] U.S. Pat. No. 11,259,871 [Non-patent literature]

[0006] [Non-Patent Document 1] Zhenliang He, Wangmeng Zuo, Meina Kan, Shiguang Shan, and Xilin Chen, “AttGAN: Facial Attribute Editing by Only Changing What You Want,” IEEE Transactions on Image Processing, 2018 [Non-Patent Document 2] Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio, “Generative Adversarial Nets”, Advances in Neural Information Processing Systems, pp. 2672-2680, 2014

[0007] The patent or patent application file contains at least one color executed drawing. Copies of this patent or patent application publication containing color drawing(s) will be provided by the U.S. Patent and Trademark Office upon request and payment of the necessary fee. [Brief description of the drawings]

[0008] [Figure 1] FIG. 1 shows a 3D graphic of the heart showing tissue condition. [Diagram 2] FIG. 13 shows sestamibi images showing perfusion within regions of the left ventricle. [Diagram 3] FIG. 1 shows images representing different tissue conditions. [Figure 4] This is an echocardiogram. [Diagram 5] FIG. 1 illustrates a cardiac artery with normal blood flow and with severely restricted blood flow. [Figure 6] FIG. 13 is a diagram showing a voltage map. [Figure 7] FIG. 1 shows an MRI image of the atrium. [Figure 8] FIG. 1 shows an MRI image representing a slice of the left ventricle. [Figure 9] FIG. 1 is a schematic diagram providing a 2D view of an example reentry circuit. [Figure 10] FIG. 1 is a flow diagram illustrating the overall processing of a TSGD system in some embodiments. [Figure 11] FIG. 2 is a block diagram illustrating components of a TSGD system and an RML system in some embodiments. [Figure 12] FIG. 13 is a flow diagram illustrating the processing of the transcription tissue state component of the TSGD system in some embodiments. [Figure 13] FIG. 13 is a flow diagram illustrating the processing of an additional occurrence site component of the TSGD system in some embodiments. [Figure 14] FIG. 1 is a flow diagram illustrating the processing of a display 3D graphics component of the TSGD system in some embodiments. [Figure 15] FIG. 2 is a flow diagram illustrating the processing of a generated RML model in an RML system in some embodiments. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Methods and systems are provided for displaying graphics showing areas of tissue status (normal or abnormal) of tissue of an organ in a subject and for indicating the origin of the source of abnormal electrical activity of the organ. The subject can be a human or an animal, and the organ can be, for example, the heart, brain, gastrointestinal tract, lungs, liver, kidneys, stomach or skeletal muscle, whose activity can be measured preferably non-invasively (e.g., from outside the subject's body) or minimally invasively. The tissue status can also represent the activity of the organ from normal to abnormal, such as electrical, metabolic, motion or perfusion. The tissue status can also be represented by other characteristics such as organ wall thickness. The tissue status can be classified as normal tissue, border zone tissue and scar tissue, for example, ranging from normal to no electrical activity, normal to no metabolic activity, normal to no perfusion, normal to limited motion, and normal to minimal wall thickness. The tissue status can be derived from scans or other measurements of the subject's organ (subject's organ). The scans may include, for example, one or more of sestamibi, positron emission tomography (PET), echo, computed tomography, voltage mapping, and magnetic resonance imaging (MRI) scans. Below, the tissue status graphic display (TSGD) system is described in the context of an organ, primarily the heart. The methods and systems of the TSGD system described below for the heart can be employed to provide graphics of other organs. The graphics provide a multi-dimensional representation of the heart, such as a two-dimensional (2D), three-dimensional (3D), or four-dimensional (4D) representation, visually indicating the state of the tissue and the site of origin of abnormal electrical activity (e.g., arrhythmia). The term "patient heart" is used to refer to the entire patient heart or a portion of the patient heart, such as the left atrium or right ventricle.

[0010] As discussed above, electrical tissue conditions can be classified as normal tissue, border zone tissue, or scar tissue, the latter two being abnormal tissue conditions. Border zone tissue is intermediate between normal tissue and scar tissue and exhibits intermediate activity between normal and scar tissue. Abnormal tissue often contains a source of abnormal electrical activity initiated by cells at the site of origin. To prevent such abnormal electrical activity, an electrophysiologist (EP) can perform an ablation procedure that targets cells near the site of origin and potentially all cells in the border zone and scar tissue. One non-invasive technique for identifying the site of origin (or source region) is described in U.S. Pat. No. 10,860,754, issued Dec. 18, 2020, and entitled "Calibration of Simulated Cardiograms," which is incorporated herein by reference.

[0011] In some embodiments, the TSGD system displays a graphic that provides information about the regions of tissue condition ranging from normal to abnormal and the site of origin of arrhythmias in the patient (subject). The EP can use the information to assist in formulating a treatment plan for the patient, such as determining where to initially place a catheter to initiate pacing to determine the target of ablation. The graphic can have a shape based on a general shape or a patient-specific shape. The shape of the patient's heart can be determined based on measurements collected from the patient (e.g., via echocardiography, CT, and anatomical mapping), which is identified based on simulations with different cardiac geometries (e.g., as described in U.S. Pat. No. 10,952,794, issued March 23, 2021, and entitled "Augmentation of Images with Source Location," which is incorporated herein by reference). Different tissue conditions can be represented by different colors, such as a range of colors from blue for normal perfusion to red for no perfusion. The TSGD system also adds an indication of the site of origin to the graphic, which can be illustrated as contours representing the probability that the site of origin is within each contour, with the center contour representing the highest probability. Probability can also be represented by a different color than that used for tissue state. Figure 1 shows a 3D graphic in which normal tissue 101, border zone tissue 102, and scar tissue 103 are illustrated in black, gray, and red, respectively. Sites of development 104 are illustrated by contour lines. (Note: if the drawing is not available in color, colors are represented in various intensities of grayscale.)

[0012] The TSGD system identifies tissue conditions (e.g., levels of perfusion) from scans of a patient's heart. The scan typically produces 2D images representing slices of the heart. Each 2D image may include a color indicative of the tissue condition of the portion of the patient's heart represented by the 2D image. The tissue condition may alternatively be identified in metadata associated with the 2D image, such as metadata including a color or other identifier of the tissue condition of each pixel of the image. The metadata may also indicate the portion of the patient's heart that the slice represents, such as the orientation and location of the heart wall. For each 2D image, the TSGD system maps that 2D image to a corresponding 3D model slice of a 3D model of the heart that represents the heart geometry and includes an indication of the location of origin. The TSGD system then adds (directly or via metadata) to that 3D model slice a designation of the tissue condition represented by the 2D image. The 3D model may also represent different sublayers of the heart wall layers. The layers of the heart include the endocardium, myocardium, and epicardium. The myocardium may have, for example, 16 sublayers that span the thickness of the myocardium. Once a 3D graphic is generated based on the 3D model, it can display the tissue condition of the selected sublayer. Once a different sublayer is selected for display (e.g., by a user), the EP can analyze the sublayer to assess the full extent (volume) of the tissue condition and site of origin. Additionally, the TSGD system can also display 2D graphics representing slices of the 3D model. Such 2D graphics represent the tissue condition and site of origin across all sublayers of the heart wall. Additionally, the TSGD system can calibrate the 3D model based on the geometry of the patient's heart as represented by the 2D image.

[0013] The TSGD system may represent the 3D model (e.g., of FIG. 1) in a Digital Imaging and Communications in Medicine (DICOM) format file (possibly as metadata) and provide the DICOM file to an external system that identifies areas of normal tissue, border areas, and scar tissue that may differ from the 3D model. The external system may overlay the identified areas on the 3D model represented by the DICOM file. Other file formats, such as stereolithography (STL) or OBJ file formats, may be used. The overlapping and non-overlapping areas of the 3D model and the areas identified by the external system may be illustrated using different colors, cross-hatching, shading, etc. For example, to distinguish between the scar tissue of the exported 3D model and the non-overlapping areas of scar tissue identified by the external system, the areas of overlapping scar tissue may be colored a dark red, and the non-overlapping areas may be illustrated in different shades of red. The external system may also adjust the heart shape of the 3D graphic and the areas and occurrences of different tissue conditions based on the heart shape identified by the external system.

[0014] In the following, we describe the TSGD system applied to various modalities for tissue status identification.

[0015] (Sestamibi) In sestamibi, a radioactive dye that emits gamma rays is injected into the bloodstream and images are collected from the patient. Measuring the emission of gamma rays allows for the quantification of perfusion through tissue and the differentiation of normal tissue, border zone, and scar tissue. FIG. 2 shows sestamibi images showing perfusion within regions of the left ventricle. Legend 203 indicates the orientation of the left ventricle, with anterior being superior and septal being inferior. The top row 201 shows perfusion during stress (e.g., during a treadmill test) and the bottom row 202 shows perfusion at rest. Legend 204 shows a color scale of perfusion, ranging from 100% complete perfusion (normal tissue) in red to 0% perfusion (scar tissue) in purple. The range corresponding to border zone tissue can be set to a range between 25% and 75%, for example. Each image in the row represents a different slice of the left ventricle. The TSGD system maps the slices onto a 3D model of the myocardium and overlays the sites of origin as shown in FIG. 1.

[0016] (PET) In PET, images are collected from a patient by injecting a radioactive dye that emits positrons into the bloodstream. By measuring the amount of positron emission, perfusion through the tissue and tissue metabolism can be quantified to indicate normal tissue, border zones, and scar tissue. FIG. 3 shows images representing different tissue states. Image 301 represents perfusion ranging from normal (top) to no or minimal perfusion (bottom). Image 302 represents metabolism as measured by fluorodeoxyglucose (FDG) uptake by cells. The top and bottom images represent no (or minimal) FDG uptake, while the images in between represent FDG uptake. Disease states (e.g., scar tissue) are indicated based on a combination of perfusion and FDG uptake. For example, as shown in the second image from the top, minimal reduction in perfusion (shown in yellow) combined with FDG uptake (shown in yellow) represents a mild disease state or border zone tissue. The TSGD system maps the images onto slices of the 3D model and overlays the site of origin as shown in FIG. 1.

[0017] (Echocardiogram) In an echocardiogram, images are collected using transthoracic, esophageal, or intracardiac imaging. These images can be analyzed to identify normal, reduced, and severely reduced motion, which correspond to normal tissue, border zone tissue, and scar tissue, respectively. Images can also be analyzed to determine the thickness of the heart wall and identify normal, reduced, and severely reduced thickness, which correspond to normal tissue, border zone tissue, and scar tissue, respectively. FIG. 4 shows an echocardiogram. Motion refers to the movement of the myocardium associated with contraction and enhancement of the heart. Given a series of images, pixels corresponding to points on the myocardium during contraction and enhancement can be analyzed to determine the distance of movement of the points as an assessment of motion. Significant, moderate, and minimal motion indicate normal, border zone, and scar tissue, respectively. Thickness indicates the thickness of the myocardium resulting from strain during contraction and enhancement. The thickness of normal tissue increases and decreases during contraction and enhancement. In contrast, the thickness of scar tissue tends not to change (or at least not change significantly) during contraction and enhancement. Given a series of images, pixels corresponding to points on the endocardium and epicardium can be analyzed from one image to the next to determine thickness. Thickness can be assessed based on points on the endocardium and nearest points on the epicardium. The TSGD system maps the assessment of normal tissue, border zones and scar tissue determined based on myocardial motion and thickness on the 3D model, overlaying the site of origin as shown in FIG. 1. Transesophageal echocardiography can also be used to collect intracardiac images as a sequence of 3D images, called 4D images, which are analyzed to determine motion and wall thickness.

[0018] (CT Imaging) In CT imaging, images are collected using a contrast agent that can be injected into the blood. Typically, the amount of contrast agent used is the minimum amount necessary to determine blood flow through the blood vessels. However, increasing the amount of contrast agent can quantify perfusion within the cardiac tissue. Once perfusion is quantified, techniques similar to those described above (e.g., sestamibi imaging) can be used to identify normal tissue, border zones, and scar tissue and map them onto a 3D model with the site of origin superimposed.

[0019] CT images acquired with normal contrast volumes can also be used to assess blood flow in cardiac arteries. Portions of cardiac arteries with little or no blood flow (e.g., significant calcium deposits) can indicate nearby scar tissue formation. The TSGD system generates a 3D model showing blood flow through portions of cardiac arteries. FIG. 5 shows cardiac arteries with normal and severely restricted blood flow. Green and red arrows point to portions of normal and severely restricted blood flow, respectively. The amount of blood flow can also be shown by overlaying cardiac arteries, for example, in red, indicating normal blood flow, gray, indicating restricted blood flow, and black, indicating severely restricted or no blood flow. The site of origin superimposed on the 3D graphic allows cardiologists to more effectively assess, for example, ablation targets.

[0020] (Voltage map) Cardiac voltage maps can be collected, for example, using a basket catheter. Figure 6 shows the voltage map. The red areas represent scar tissue and the yellow and blue areas represent border zone tissue. The TSGD system can overlay the site of origin onto the 2D voltage map. The TSGD system can also map the voltage map onto a 3D graphic with the site of origin superimposed, as shown in Figure 1.

[0021] (MRI) In MRI imaging, images can be collected from a patient with a contrast agent injected into the patient's blood to quantify cardiac tissue perfusion. Figure 7 is an MRI image of the atrium. Blue indicates normal tissue and green indicates scar tissue (e.g., fibrosis). Figure 8 is a slice MRI image of the left ventricle. The yellow dotted lines represent areas of reduced perfusion, indicating scar tissue. The TSGD system can generate 3D models in a similar manner as described above for sestamibi imaging.

[0022] (Reentry Machine Learning System) A reentry machine learning (RML) system is provided that identifies the entry site, isthmus, and / or exit site of the reentry circuit of a reentry arrhythmia based on at least the tissue condition of the reentry circuit. The exit site can be targeted for ablation to stop the function of the reentry circuit. The site of occurrence of electrical activity that results in activation of the reentry circuit may be within the reentry circuit or outside the reentry circuit. FIG. 9 is a schematic diagram providing a 2D illustration of an exemplary reentry circuit. A common pathway (CP) 915 (e.g., isthmus) between scar tissue is a pathway from the CP entry site 910 to the CP exit site 901. The inner loops 901, 903, 906, and 910 lead from the CP exit site to the CP entry site. The outer loops 901, 921, 922, 925, 930, and 910 lead from the CP exit site to the CP entry site. Channels C911, E912, and H913 are dead-end channels where the action potential enters and terminates in the dead-end channel. The RML system can generate a 3D model of the reentry circuit derived from a 3D model of the heart that includes a specification of the tissue state generated by the TSGD system or obtained from other systems.

[0023] In some embodiments, the RML system trains the RML model with training data that specifies characteristics of the reentry circuit including entry sites, exit sites, isthmus characteristics (e.g., pathways), and tissue characteristics including shape (e.g., 3D), site, and properties (e.g., perfusion, electrical, motion, etc.). The training data can be based on clinical data and / or simulated data collected from a subject. Clinical data can be collected by analyzing tissue condition characteristics (e.g., perfusion or electrical activity) of the subject's reentry circuit. Simulated data can be generated by simulating cardiac electrical activity assuming certain tissue properties.

[0024] Simulations of cardiac electrical activity are described in U.S. Patent No. 11,259,871, issued March 1, 2022, and entitled "Identify Ablation Pattern for Use in an Ablation," which is incorporated herein by reference. The simulations use a 3D mesh with vertices representing sites within the heart and having electrical properties such as conduction velocity and action potential. Scar tissue can be represented as vertices having properties of no electrical activity, and border zone tissue can be represented as vertices having properties of limited electrical activity. Each simulation can have parameters that specify one or more regions of tissue (e.g., scar tissue and border zone tissue) that may appear to function as reentry circuits during the simulation. The regions used in the simulation can be derived from a library of 2D images that show tissue conditions (e.g., reentry circuit conditions) collected from a patient. The regions can be augmented with additional regions that modify the regions derived from the 2D images. Alternatively or additionally, the RML system can generate the regions using a rule-based algorithm that specifies the properties of the regions that may function as reentry circuits. The characteristics are the geometry (e.g., 3D shape), location, and electrical activity of the border zone and scar tissue as in Figure 9. A cardiac cycle (e.g., fibrillation cycle) resulting from a reentry circuit represents the electrical activity from the exit site to the entry site and through the isthmus to the exit site. Each simulation can simulate the electrical activity for a certain period of time or until, for example, fibrillation stabilizes (i.e., beat-to-beat consistency of a dominant arrhythmia source localized to a particular region within the heart).

[0025] Once the simulation is complete, the RML system identifies loops near the scar tissue based on the flow of action potentials that loop back on themselves during one cycle. The entry site of the loop can be identified based on analysis of conduction velocity. Isotropic properties of conduction velocity based on along-fiber conduction velocity and across-fiber conduction velocity aid in identifying the entry site. The RML system identifies the exit site based on analysis of an electrocardiogram (e.g., a vector electrocardiogram (VCG)) that is dynamically generated from a series of simulated electrical activity represented by calculated values ​​such as action potentials at the vertices of a 3D mesh. Once the RML system identifies the onset of depolarization near an entry site, it analyzes the electrical activity to identify where the depolarization was initiated. The RML system can generate probabilities of being an exit site for multiple locations. An isthmus is a pathway (e.g., between scar tissue) that is in the direction of action potential flow along the loop from the entry site to the exit site.

[0026] To identify the patient's exit site (and possibly entry site and isthmus), 2D images are collected from the patient showing perfusion, motion, electrical activity, etc. This 2D image or a 3D image derived from the 2D image and / or features derived from the 2D image such as areas of scar tissue are input into an RML model, which outputs the exit site and possibly the entry site and / or isthmus. The RML system can then display the reentry circuit as a 2D graphic (e.g., FIG. 9) or as a 3D graphic similar to the 3D graphic based on the 3D model generated by the TSGD system. To provide the entry and isthmus features, a separate model can be trained to input the exit site information and scar tissue information and output a 2D or 3D model of the reentry circuit with entry sites and isthmus defined.

[0027] The RML system can show one or more exit sites (and corresponding loop and entry sites) on a 2D, 3D or 4D graphic of the heart. The various potential exit sites for the reentry circuit can be shown with color intensities representing the probability of each exit site.

[0028] A computer system (e.g., a network node or collection of network nodes) on which the TSGD system, the RML system, and other described systems may be implemented may include a central processing unit, input devices, output devices (e.g., display devices and speakers), storage devices (e.g., memory and disk drives), network interfaces, graphics processing units, communication links (e.g., Ethernet, Wi-Fi, cellular, and Bluetooth), global positioning system devices, and the like. Input devices may include keyboards, pointing devices, touch screens, gesture recognition devices (e.g., for air gestures), head and eye tracking devices, microphones for voice recognition, and the like. Computer systems may include high performance computer systems, cloud-based computer systems, client computer systems that interact with cloud-based computer systems, desktop computers, laptops, tablets, e-readers, personal digital assistants, smartphones, gaming devices, servers, and the like. Computer systems may access computer readable media, including computer readable storage media and data transmission media. A computer readable storage medium is a tangible storage means that does not include a transitory propagating signal. Examples of computer-readable storage media include memory and other storage, such as primary memory, cache memory, secondary memory (such as DVDs). The computer-readable storage media may have computer-executable instructions or logic stored thereon or encoded thereon that implement the TSGD system, the RML system, and other described systems. The data transmission media is used to transmit data via ephemeral propagating signals or carrier waves (e.g., electromagnetic) over wired or wireless connections. The computer system may include a secure crypto-processor as part of the central processing unit to generate and securely store keys, and to encrypt and decrypt data using the keys.

[0029] The TSGD system, the RML system, and other described systems may be described in the general context of computer-executable instructions, such as program modules and components, executed by one or more computers, processors, or other devices. Program modules or components include routines, programs, objects, data structures, etc. that perform the tasks or implement data types of the TSGD system, the RML system, and other described systems. Typically, the functionality of the program modules may be combined or distributed as desired. Aspects of the TSGD system, the RML system, and other described systems may be implemented in hardware using, for example, application specific integrated circuits (ASICs) or field programmable gate arrays (FPGAs).

[0030] The machine learning (ML) model may be any of a variety or combination of supervised or unsupervised machine learning models, including neural networks such as fully connected neural networks, convolutional neural networks, recurrent neural networks, or autoencoder neural networks, or restricted Boltzmann machines, support vector machines, Bayesian classifiers, k-means clustering techniques, etc. If the machine learning model is a deep neural network, the learning result is a set of weights for the activation function of the deep neural network. Support vector machines operate by finding a hypersurface in the space of possible inputs. The hypersurface attempts to separate positive and negative examples by maximizing the distance between the positive and negative examples closest to the hypersurface. This step allows for the correct classification of data that is similar but not identical to the training data. Machine learning models can generate discrete domain values ​​(e.g., classification), probabilities, and / or continuous domain values ​​(e.g., regression values).

[0031] Various techniques such as adaptive boosting can be used to train a support vector machine. Adaptive boosting is an iterative process of running multiple tests on a set of training data. Adaptive boosting transforms a weak learning algorithm (one that performs at a level only slightly better than chance) into a strong learning algorithm (one that exhibits a low error rate). The weak learning algorithm is run on different subsets of the training data. The algorithm focuses more and more on examples where its predecessor is more likely to show mistakes. The algorithm corrects for mistakes made by the previous weak learner. The algorithm is adaptive because it adapts to the error rate of its predecessor. Adaptive boosting combines rough and moderately inaccurate heuristics to produce a high-performance algorithm. Adaptive boosting combines the results of tests run separately into one accurate classifier. Adaptive boosting sometimes uses weak classifiers, which are one-way trees with only two leaf nodes.

[0032] A neural network model has three main components: architecture, cost function, and search algorithm. The architecture defines the functional form that relates inputs and outputs (in terms of the network's topology, unit connectivity, and activation functions). Searching the weight space for a set of weights that minimizes the objective function is the learning process. In one embodiment, the classification system can use radial basis function (RBF) networks and standard gradient descent as the search technique.

[0033] A convolutional neural network (CNN) has multiple layers, such as convolutional layers, rectified linear unit (ReLU) layers, pooling layers, fully connected (FC) layers, etc. More complex CNNs may have multiple convolutional layers, ReLU layers, pooling layers, and FC layers.

[0034] A convolutional layer includes multiple filters (also called kernels or activation functions). A filter inputs, for example, a convolutional window of an image, applies weights to each pixel in the convolutional window, and outputs an activation value for that convolutional window. For example, if a still image is 256x256 pixels, the convolutional window is 8x8 pixels. A filter can apply different weights to each of the 64 pixels in the convolutional window to generate activation values, also called feature values. A convolutional layer can include a node (also called a neuron) for each pixel of the image, assuming one stride with appropriate padding, for each filter. Each node outputs a feature value based on a set of learned filter weights.

[0035] The ReLU layer may have a node that generates a feature value for each node in the convolutional layer. The generated feature values ​​form a ReLU feature map. The ReLU layer applies a filter to each feature value of the convolutional feature map to generate a feature value of the ReLU feature map. For example, a filter such as max(0, activation value) may be used to prevent the feature value of the ReLU feature map from becoming negative.

[0036] The pooling layer can be used to reduce the size of the ReLU feature map by downsampling the ReLU feature map to form a pooled feature map. The pooling layer includes a pooling function that inputs the feature values ​​of the ReLU feature map and outputs feature values.

[0037] The FC layer contains several nodes, each connected to each feature value in the pooling feature map.

[0038] Generative Adversarial Networks (GANs) or attributes (attGANs) can also be used. attGANs use GANs to train generative models. (See Zhenliang He, Wangmeng Zuo, Meina Kan, Shiguang Shan, and Xilin Chen, "AttGAN: Facial Attribute Editing by Only Changing What You Want," IEEE Transactions on Image Processing, 2018, and Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio, "Generative Adversarial Nets," Advances in Neural Information Processing Systems, pp. 2672-2680, 2014, which are incorporated herein by reference.) attGANs include a generator, a discriminator, and an attGAN classifier, and are trained using training data that includes input images of objects and input attribute values ​​for each object. The generator includes a generator encoder and a generator decoder. The generator encoder inputs an input image and is trained to generate a latent vector of latent variables that represent the input image. The generator decoder inputs the latent vector of the input image and the input attribute values. The attGAN classifier inputs an image and generates a prediction for the attribute values. The attGAN is trained to generate modified images that represent the input image modified based on the attribute values. The generator encoder and generator decoder form a generator model.

[0039] FIG. 10 is a flow diagram illustrating the overall processing of the TSGD system in some embodiments. A TSGD controller 1000 controls the generation of a 3D model and the display of the 3D model as a 3D graphic with tissue status and site of origin information. In block 1001, the controller receives an electrogram of the electrical activity of the organ. The electrogram may be collected, for example, using a 12-lead electrocardiograph. In block 1002, the component receives a tissue status image, which may be collected using a scanning technique such as a sestamibi scan. In block 1003, the component generates a 3D model representing the patient's organ. In block 1004, the component transfers tissue status from the tissue status image to the 3D model to augment the 3D model. In block 1005, the component adds site of origin information to the 3D model. In block 1006, the component displays a 3D graphic showing the tissue status and site of origin of the organ based on the 3D model. The component then completes.

[0040] 11 is a block diagram illustrating components of the TSGD system and the RML system in some embodiments. The TSGD system 1110 includes a TSGD controller 1111, a transfer tissue state component 1112, an add site of origin component 1113, a display 3D graphics component 1114, and a mapping system 1115. As described above, the TSGD controller controls the overall operation of the TSGD system. The transfer tissue state component receives 2D images showing tissue states and transfers the tissue states to a 3D model. The add site of origin component receives an electrocardiogram, accesses a mapping system to identify site of origin information, and adds the site of origin information to the 3D model to augment the 3D model. The display 3D graphics component generates a 3D graphic from the 3D model with the tissue state and site of origin information, and controls the display of the 3D graphic.

[0041] The RML system 1120 includes a generate RML model component 1121, an apply RML model component 1122, an RML model weights data store 1123, and an RML model training data store 1124. The generate RML model component generates an RML model using training data and stores the learned RML model weights in the RML model weights data store. The apply RML model component inputs the exit site and scar tissue information and applies the RML model to the exit site and scar tissue using the RML model weights to determine the corresponding entry site and isthmus. Various components of the TSGD system and the RML system may be provided by different computer systems. For example, the mapping system may be implemented on a cloud-based computer system and the display 3D graphics component may be implemented on a client computer system. When implemented on different computer systems, the computer systems may send data to and receive data from each other. For example, a client computer system can transmit electrograms, such as an electrocardiogram (e.g., an electrocardiogram (ECG) or a vectorcardiogram (VCG)), an electrogastrogram (EGG), or an electroencephalogram (EEG), to a cloud-based mapping system and receive site of origin information from the cloud-based mapping component.

[0042] 12 is a flow diagram illustrating the processing of the transition tissue state component of the TSGD system in some embodiments. The transition tissue state component 1200 inputs a 3D model and a 2D image showing the tissue state and transitions the tissue state to the 3D model. In block 1201, the component selects the next slice of the 2D image and the corresponding slice of the 3D model. In decision block 1202, if all slices have already been selected, the component is done, otherwise the component proceeds to block 1203. In block 1203, the component selects the next sublayer of the selected slice. In decision block 1204, if all sublayers have already been selected, the component loops to block 1201 to select the next sublayer of the selected slice, otherwise the component proceeds to block 1205. In block 1205, the component transitions the tissue state to the selected sublayer of the selected slice of the 3D model and then loops to block 1203 to select the next sublayer.

[0043] 13 is a flow diagram illustrating the processing of the Add Site of Origin component of the TSGD system in some embodiments. The Add Site of Origin component is invoked with an indication of an ECG and a 3D model to transfer the site of origin information to the 3D model. In block 1301, the component sends the ECG to a mapping system to identify the site of origin information. In block 1302, the component receives the site of origin information. In block 1303, the component selects a slice of the 3D model. In decision block 1304, if all slices have already been selected, the component is done, otherwise the component proceeds to block 1305. In block 1305, the component selects the next sublayer of the selected slice. In decision block 1306, if all sublayers have already been selected, the component loops to block 1303 to select the next slice, otherwise the component proceeds to block 1307. In decision block 1307, if the site of origin information overlaps with the selected sublayer in the selected slice, the component continues at block 1308, else the component loops to block 1305 to select the next sublayer. In block 1308, the component marks the portion of the 3D model represented by the selected slice and the selected sublayer with the site of origin information.

[0044] FIG. 14 is a flow diagram illustrating the processing of the display 3D graphics component of the TSG system in some embodiments. The display 3D graphics component 1400 is passed a 3D model and controls the display of a 3D graphic generated from the 3D model. In block 1401, the component generates a 3D graphic based on the 3D model. In block 1402, the component outputs a graphic, such as a 3D graphic or a 2D slice. In block 1403, the component receives a selection of an action, which is to end the display of the 3D graphic, to rotate the 3D graphic, to display a particular sublayer, and to display a particular slice. In decision block 1404, if the action is to rotate the 3D graphic, the component proceeds to block 1405 to generate a rotated 3D graphic, otherwise the component proceeds to block 1406. In block 1406, if the action is to display a particular sublayer, the component proceeds to block 1407 to generate a 3D graphic based on that sublayer, otherwise the component proceeds to block 1408. In decision block 1408, if the action is to display a particular slice, the component continues to block 1409 to generate a 2D slice from the 3D model, else the component completes. After performing the processing in blocks 1405, 1407, and 1409, the component loops to block 1402 to output the graphic.

[0045] FIG. 15 is a flow diagram illustrating the processing of a generative RML model of an RML system in some embodiments. The generative RML model component 1500 trains an RML model based on training data in an RML model training data store. The training data includes training instances that may specify reentry circuits or non-reentry scar tissue with an indication of the exit site of the reentry circuit or no exit site of the non-reentry scar tissue. Alternatively, the training data may only have training instances of reentry circuits. If both reentry and non-reentry scar tissue are used, one ML model may be trained to distinguish between reentry circuits and non-reentry scar tissue, and another ML model may be trained to indicate the exit site of the reentry circuit. And another ML model may be trained to identify entry sites and isthmuses given tissue features and exit sites of the reentry circuit. Each ML model is trained with training data that includes output data to be identified as labels and input data used to identify the output data as feature vectors. In block 1501, the component accesses the training data of the RML model. In block 1502, the component selects the next training instance of the RML training data. In block 1503, the component generates a feature vector containing tissue properties (e.g., perfusion or electrical properties) in the 3D model of the training instance. In block 1504, the component labels the feature vector with the exit site and possibly the entry site and isthmus information of the training instance. In decision block 1505, if all training instances have been selected, the component continues at block 1506, else the component loops to block 1502 to select the next training instance. In block 1506, the component trains an RML model using the labeled feature vector, stores the learned weights in an RML model weights data store, and completes.

[0046] The following paragraphs describe various aspects of the TSGD and RML systems. An implementation of the systems may employ any combination of the aspects. The processes described below may be performed by a computer system having a processor executing computer-executable instructions stored on a computer-readable storage medium that implements the system.

[0047] In some aspects, the techniques described herein relate to a method, performed by one or more computer systems, of augmenting a three-dimensional (3D) model of the heart to indicate tissue status, the method including the steps of: accessing a 3D model of the heart; accessing two-dimensional (2D) images of tissue status slices of the heart, the tissue status slices having cardiac tissue status information; accessing arrhythmia site information; augmenting the 3D model with a representation of the site of occurrence based on the site of occurrence information; for each of a plurality of tissue status slices of the heart, augmenting a 3D model slice of the 3D model corresponding to the tissue status slice with a representation of the cardiac tissue status represented by the tissue status information of the tissue status slice; and displaying a representation of the 3D model showing the site of occurrence of the arrhythmia and the cardiac tissue status. In some aspects, the techniques described herein relate to a method in which the displayed representation of the 3D model is a 3D graphic. In some aspects, the techniques described herein relate to a method in which the displayed representation of the 3D model is a slice of the 3D model. In some aspects, the techniques described herein relate to a method, where the 3D model includes a plurality of 3D model sublayers of a layer of a heart wall of the 3D model, the augmentation of the 3D model enhances the plurality of 3D model sublayers with an indication of the site of origin represented by the site of origin information, and the augmentation of the 3D model slice enhances the plurality of 3D model sublayers with an indication of the tissue state represented by the tissue state information. In some aspects, the techniques described herein relate to a method, where the layer is an endocardium, myocardium, or epicardium of the heart wall. In some aspects, the techniques described herein relate to a method, where the layer further includes receiving a selection of the 3D model sublayer, and the representation of the 3D model is a 3D graphic showing the site of origin and the tissue state of the selected 3D model sublayer. In some aspects, the techniques described herein relate to a method, where the augmentation of the 3D model sublayer is performed dynamically as the sublayer is selected. In some aspects, the techniques described herein relate to a method, where the 2D image is obtained from a sestamibi scan of the heart.In some embodiments, the techniques described herein relate to methods where the 2D image is derived from a positron emission tomography scan of the heart. In some embodiments, the techniques described herein relate to methods where the 2D image is derived from an echocardiography scan of the heart. In some embodiments, the techniques described herein relate to methods where the 2D image is derived from a computed tomography scan of the heart. In some embodiments, the techniques described herein relate to methods where the 2D image is derived from a voltage map of the heart. In some embodiments, the techniques described herein relate to methods where the 2D image is derived from a magnetic resonance imaging scan of the heart. In some embodiments, the techniques described herein relate to methods where the 2D image is a slice of a 3D image of a scan of the heart. In some embodiments, the techniques described herein relate to methods where the tissue state of the 2D slice is based on a scan indicative of cardiac perfusion within the heart. In some embodiments, the techniques described herein relate to methods where the tissue state of the 2D slice is based on a scan indicative of cardiac wall motion of the heart. In some embodiments, the techniques described herein relate to methods where the tissue state is indicative of normal tissue, border zone tissue and scar tissue. In some embodiments, the techniques described herein relate to methods where the tissue state is based on electrical, metabolic and / or perfusion activity of the heart. In some aspects, the technology described herein relates to a method further comprising, for each of a plurality of 3D models of a four-dimensional (4D) model of the heart, augmenting the 3D model to indicate a site of origin and tissue status of the heart. In some aspects, the technology described herein relates to a method, wherein the 4D model represents cardiac wall motion of the heart. In some aspects, the technology described herein relates to a method, wherein the step of accessing site of origin information comprises accessing an electrocardiogram and identifying the site of origin information based on a mapping that respectively maps the electrocardiogram to the site of origin information. In some aspects, the technology described herein relates to a method, wherein a 3D model is generated from a 2D image of the heart based on a mapping that respectively maps the 2D image to the 3D image. In some aspects, the technology described herein relates to a method, wherein the 3D model is generated based on an anatomical mapping of the heart.In some aspects, the technology described herein relates to a method in which a 3D model is generated based on a scan of the heart, in some aspects, the technology described herein relates to a method further comprising displaying a 3D model slice graphic including an indication of the origin and tissue state of the 3D model slice, in some aspects, the technology described herein relates to a method in which the augmentation of the 3D model uses different colors to indicate different tissue states.

[0048] In some aspects, the technology described herein relates to one or more computer systems for augmenting a three-dimensional (3D) model of an organ to indicate tissue conditions of the organ, the one or more computer systems comprising one or more computer readable storage media storing a model of the organ including an indication of sites of occurrence of abnormalities of the organ, a tissue state representation of the tissue state of the organ based on a scan of the organ, and computer executable instructions for controlling the one or more computer systems to augment the model with tissue conditions of the organ based on the tissue state representation and output a representation of the model indicating sites of occurrence and tissue conditions of the organ, the one or more computer systems further comprising one or more processors for controlling the one or more computer systems to execute the one or more computer executable instructions. In some aspects, the technology described herein relates to one or more computer systems, a first computer system storing instructions for augmenting and a second computer system storing instructions for outputting. In some aspects, the technology described herein relates to one or more computer systems, where a first computer system includes instructions for receiving a tissue status representation, providing the tissue status representation to a second computer system, receiving an output representation from the second computer system, and displaying the output representation. In some aspects, the technology described herein relates to one or more computer systems, where the second computer system is a cloud-based system. In some aspects, the technology described herein relates to one or more computer systems, where the organ is selected from the group consisting of heart, brain, digestive tract, lung, liver, kidney, stomach, and muscle. In some aspects, the technology described herein relates to one or more computer systems, where the scan is a non-invasive scan.

[0049] In some aspects, the technology described herein relates to one or more computer systems for generating a reentry machine learning (RML) model for identifying characteristics of a cardiac reentry circuit, the one or more computer systems comprising one or more computer readable storage media, the one or more computer readable storage media storing computer executable instructions for controlling the one or more computer systems to access training data including descriptions of features of the reentry circuit, and for each description, computer executable instructions for extracting one or more features from the description, computer executable instructions for extracting one or more labels from the description, computer executable instructions for labeling the one or more features with one or more labels, and computer executable instructions for training the RML model using the labeled features, the one or more computer systems further comprising one or more processors for controlling the one or more computer systems to execute the one or more computer executable instructions. In some aspects, the technology described herein relates to one or more computer systems, the one or more features including tissue state information of the reentry circuit, and the one or more labels including an exit site. In some aspects, the technology described herein relates to one or more computer systems in which the features are images of areas of scar tissue.In some aspects, the technology described herein relates to one or more computer systems for identifying an exit site of a cardiac reentry circuit, the one or more computer systems comprising one or more computer readable storage media, the one or more computer readable storage media storing computer executable instructions for controlling the one or more computer systems to: access characteristics of scar tissue of a subject's reentry circuit; access a reentry machine learning (RML) model for identifying an exit site of the reentry circuit; train the RML model using training data, for each of a plurality of reentry circuits, including information related to the scar tissue of the reentry circuit labeled with information related to the exit site of the reentry circuit; apply the RML model to the characteristics of the subject's scar tissue to identify subject information related to the exit site of the reentry circuit; and output a display of the identified subject information; and the one or more computer systems further comprising one or more processors for controlling the one or more computer systems to execute the one or more computer executable instructions.

[0050] Although the subject matter of the present invention has been described in terms of specific structural features and / or acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are disclosed as example forms of implementing the claims. [Explanation of symbols]

[0051] 101 Normal tissue 102 Boundary Area Organization 103 Scar tissue 104 Location

Claims

1. 1. One or more computer systems for generating a reentry machine learning (RML) model for characterizing a cardiac reentry circuit, comprising: one or more computer-readable storage media; the one or more computer-readable storage media: storing computer-executable instructions for controlling the one or more computer systems to access training data including a description of characteristics of the reentry circuit; the characteristics include a common pathway specification from an entry site to an exit site, and tissue characteristics including one or more of perfusion, electrical, and kinetic characteristics; the training data is generated by, for each of a plurality of simulated tissue properties, performing a simulation of electrical activity of a heart having the simulated tissue properties and identifying a simulated reentry circuit from the simulated electrical activity, each simulated reentry circuit having a simulated entry site, a simulated exit site, and a simulated common pathway; For each description of the characteristics of a reentrant circuit, computer-executable instructions for extracting one or more features from the description, including tissue characteristics; computer-executable instructions for extracting from said description one or more labels comprising a specification of a common path of said description; computer-executable instructions for labeling the one or more features with the one or more labels; computer-executable instructions for training the RML model using the labeled features; Store the one or more computer systems further comprising: One or more computer systems comprising one or more processors for controlling said one or more computer systems to execute one or more of said computer-executable instructions.

2. 10. The one or more computer systems of claim 1, wherein the one or more features include tissue status information of a reentry circuit and the one or more labels include an exit site.

3. The one or more computer systems of claim 1 , wherein the features are images of areas of scar tissue.

4. The computer-executable instructions further include, for each simulated tissue property: performing a simulation of cardiac electrical activity based on the simulated tissue properties; characterizing a conduction velocity loop in the vicinity of the scar tissue during a simulation cycle; identifying a simulated entry site and a simulated exit site within the identified loop; identifying a simulated common pathway based on a direction of action potential flow represented by the simulated electrical activity from the simulated entry site to the simulated exit site; 2. The one or more computer systems according to claim 1, wherein the one or more computer systems are controlled so that

5. The computer-executable instructions further include: generating a simulated electrocardiogram based on the simulated electrical activity of the simulated cycle; identifying the simulated exit site based on an analysis of the simulated electrocardiogram; identifying the simulated entry site based on isotropic characteristics of conduction velocity; 5. The one or more computer systems according to claim 4, wherein the one or more computer systems are controlled so that

6. The computer-executable instructions further include: identifying a simulated loop of a simulated flow of action potentials exhibited by the simulated electrical activity; identifying the simulated entry site based on isotropic characteristics of conduction velocity; 5. The one or more computer systems according to claim 4, wherein the one or more computer systems are controlled so that

7. The one or more computer systems of claim 1 , wherein the tissue characteristics include indications of normal tissue, scar tissue, and border zone tissue.

8. The computer-executable instructions further include: accessing patient characteristics including one or more of perfusion, electrical, and motion characteristics; applying the trained RML model to the patient characteristics to identify common pathway specifications for the patient reentry circuit; 2. The one or more computer systems according to claim 1, wherein the one or more computer systems are controlled so that

9. 10. The one or more computer systems of claim 1, wherein the computer-executable instructions further control the one or more computer systems to display a three-dimensional (3D) graphic of the heart including a representation of a patient reentry circuit.

10. 1. One or more computer systems for identifying entry sites, common pathways, and exit sites of cardiac reentry circuits, comprising: one or more computer-readable storage media; the one or more computer-readable storage media: accessing subject tissue characteristics of a subject reentry circuit of the subject, the subject tissue characteristics including scar tissue and border zone tissue characteristics; accessing a reentry machine learning (RML) model to identify entry sites, common pathways, and exit sites of the reentry circuits, the RML model being trained using training data including, for each of a plurality of simulated reentry circuits, simulated scar tissue and simulated boundary zone tissue of the simulated reentry circuits labeled with information regarding the simulated entry sites, simulated common pathways, and simulated exit sites of the simulated reentry circuits; applying the RML model to the subject's tissue properties to identify subject information regarding entry sites, common pathways, and exit sites of the subject's reentry circuit; outputting a display of the identified subject information; storing computer-executable instructions for controlling the one or more computer systems to perform the one or more computer systems further comprising: one or more processors for controlling the one or more computer systems to execute the one or more computer-executable instructions; One or more computer systems.

11. The computer-executable instructions further include, for each of the plurality of simulated tissue properties: performing a simulation of cardiac electrical activity based on the simulated tissue properties; Characterize the conduction velocity loop during the cycle, generating a simulated electrocardiogram based on the simulated electrical activity of said cycle; identifying the simulated exit site based on an analysis of the simulated electrocardiogram; Identifying simulated entry sites based on isotropic properties of conduction velocity; identifying a simulated common pathway based on a direction of action potential flow represented by the simulated electrical activity from the simulated entry site to the simulated exit site; The one or more computer systems according to claim 10, wherein the one or more computer systems are controlled so that

12. The one or more computer systems of claim 11 , wherein the simulated tissue properties include representations of normal tissue, scar tissue, and border zone tissue.

13. 12. The one or more computer systems of claim 11, wherein the computer-executable instructions further control the one or more computer systems to display a three-dimensional (3D) graphic of the heart including a representation of the subject reentry circuit.

14. 1. A method executed by one or more computer systems for simulating electrical activity of a cardiac reentry circuit, comprising: generating a plurality of sets of simulation parameters each specifying a region of normal tissue, scar tissue, and border zone tissue; For each set of simulation parameters simulating the electrical activity of a heart having regions of scar tissue and border zone tissue specified by said set of simulation parameters; When the simulated arrhythmia stabilized, generating a simulated electrocardiogram based on the simulated electrical activity during a simulated cardiac cycle; analyzing the simulated electrical activity to identify loops near scar tissue based on a stream of simulated action potentials that loop back on themselves during the simulated cardiac cycle; identifying an entry site of the loop based on a simulated conduction velocity; identifying an exit site of the loop based on a simulated electrocardiogram; storing a mapping of one or more of simulation parameters to one or more of the identified entry site, the identified loop, and the identified exit site; A method comprising:

15. The method of claim 14 , further comprising identifying a common pathway based on the identified entry site, the identified loop, and the identified exit site.

16. The method of claim 14 , wherein the plurality of regions of scar tissue and border zone tissue are derived from images acquired from the patient.

17. 15. The method of claim 14, wherein the multiple regions of scar tissue and border zone tissue are generated based on rules that specify the characteristics of the regions that can function as reentry circuits.

18. The method of claim 14 , wherein the simulation is performed until the arrhythmia stabilizes.

19. 15. The method of claim 14, wherein identifying the exit site comprises applying a machine learning model to the simulated electrocardiogram to identify the exit site.

20. training a reentry machine learning model using a feature vector comprising features derived from the image of the reentry circuit; 15. The method of claim 14, wherein each feature vector has labels based on entry sites, common pathways, and exit sites identified based on simulations based on scar tissue and border zone tissue derived from images.

21. training a reentry machine learning model using a feature vector with features derived from the specification of scar tissue and border zone tissue in the reentry circuit; 15. The method of claim 14, wherein each feature vector has a label based on one or more of entry sites, common pathways, and exit sites identified based on simulations based on specifications of scar tissue and border zone tissue.

22. 1. A method executed by one or more computer systems for identifying reentry circuit characteristics of a cardiac reentry circuit, comprising: accessing a subject image of a subject reentry circuit of the subject; accessing a reentry circuit machine learning (RML) model to identify reentry circuit characteristics based on tissue characteristics for scar tissue and border zone tissue identified from the subject image, the RML model being trained using training data including, for each of a plurality of simulated reentry circuits, simulated tissue characteristics for the simulated scar tissue and simulated border zone tissue of the simulated reentry circuit labeled with the simulated reentry circuit characteristics of the simulated reentry circuit; applying the RML circuit model to tissue characteristics of the subject to identify subject reentry circuit characteristics for the subject's reentry circuit; outputting a representation of the identified reentry circuit characteristics; A method comprising:

23. 23. The method of claim 22, wherein said outputting comprises outputting a three-dimensional (3D) graphic of the heart representing the identified subject reentry circuit.

24. For each of multiple specifications of normal tissue, border zone tissue, and scar tissue of the heart, simulating electrical activity of the heart based on the specifications, the simulated electrical activity including a simulated action potential and a simulated conduction velocity; When the simulated arrhythmia stabilized, generating a simulated electrocardiogram based on the simulated electrical activity during a simulated cardiac cycle; analyzing the simulated electrical activity to identify loops near scar tissue based on a stream of simulated action potentials that loop back on themselves during the simulated cardiac cycle; identifying an entry site of the loop based on a simulated conduction velocity; identifying an exit site of the loop based on a simulated electrocardiogram; storing a mapping of specifications to one or more of the identified entry site, the identified loop, and the identified exit site; 23. The method of claim 22, further comprising:

Images