Arrhythmia source location assessment

Abstract

A method for treating a patient having an arrhythmia is provided. The method identifies an inferred source location of an arrhythmia based on a patient cardiogram along with an initial source location score indicating confidence the inferred source location is the source of the arrhythmia. The method identifies the cardiac tissue state of cardiac tissue that is proximal to the inferred source location. The method then generates a refined source location score that is a refinement of the initial source location score. The refined source location score factors in the suitability of the cardiac tissue for originating and / or sustaining an arrhythmia as indicated by the cardiac tissue state. An ablation procedure is performed on the patient to treat the arrhythmia at an ablation location that, depending on the refined source location score, may be different from the inferred source location.

Description

Attorney Docket No.: 129292.8042. WOOOARRHYTHMIA SOURCE LOCATION ASSESSMENTCROSS-REFERENCE TO RELATED APPLICATION(S)

[0001] This application claims priority to U.S. Provisional Application No.63 / 777,442, filed on March 25, 2025, entitled “ ARRHYTHMIA SOURCE LOCATION ASSESSMENT,” which is incorporated herein by reference.BACKGROUND

[0002] Many heart disorders can cause symptoms, morbidity (e.g., syncope or stroke), and mortality. Common heart disorders caused by arrhythmias include inappropriate sinus tachycardia (1ST), 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 complexes (PVCs), premature atrial complexes (PACs), atrioventricular nodal reentrant tachycardia (AVNRT), atrioventricular reentrant tachycardia (AVRT), permanent junctional reciprocating tachycardia (PJRT), and junctional tachycardia (JT). The sources of arrhythmias may include electrical rotors (e.g., VF), recurring electrical focal sources (e.g., AT), anatomically based reentry (e.g., VT), and so on. These sources are important drivers of sustained or clinically significant episodes. Arrhythmias can be treated with ablation using different technologies— including radiofrequency energy ablation, cryoablation, ultrasound ablation, laser ablation, external radiation sources, directed gene therapy, and so on — by targeting the source of the heart disorder. Since the sources of heart disorders and their locations (which may be referred to as source location or site of origin) vary from patient to patient, even for common heart disorders, targeted therapies require the source location of the arrhythmia to be identified.

[0003] Unfortunately, traditional methods for reliably identifying the sources and their source locations of a heart disorder can be complex, cumbersome, and expensive. For example, one method uses a multi-electrode basket catheter that is inserted intravascularly into the heart (e.g., left ventricle) to collect measurements of the electrical activity of the heart during an induced episode, such as VF. These measurements may be analyzed to help identify a source location. However,Attorney Docket No.: 129292.8042. WOOOelectrophysiology catheters are expensive, typically single use, and may lead to serious complications such as cardiac perforation and tamponade.

[0004] Another method uses an exterior body surface vest with electrodes to collect measurements from the patient’s torso, which are analyzed to help identify an arrhythmia source location. These vests are expensive, complex, and difficult to manufacture, and may interfere with the placement of defibrillator pads needed during induced VF episodes. In addition, analysis using such vests typically requires a computed tomography (CT) scan. Moreover, body surface vests are unable to sense electrical activity in the interventricular and interatrial septa, where approximately 20% of arrhythmia sources may occur.

[0005] Some techniques are available to determine non-invasively the source location of an arrhythmia. One such technique, referred to as a mapping system, runs simulations of electrical activity in hearts with varying simulated cardiac characteristics. These characteristics may include cardiac geometry (e.g., atrial shape) and electrophysiological properties (e.g., action potential, conduction velocity), which in turn may reflect different types of cardiac tissue, such as normal, borderzone, and scar tissue.

[0006] Each simulation assumes a source location of an arrhythmia and is run until an arrhythmia stabilizes or for a predetermined duration. For each simulation that results in an arrhythmia, a simulated cardiogram (e.g., electrocardiogram or vectorcardiogram) is generated based on the simulated electrical activity and is associated with the assumed source location. To build a comprehensive library mapping simulated cardiograms to source locations, millions of simulations may be run to encompass a broad range of cardiac characteristics and source locations. The mapping system may generate a mapping for each cycle of a simulated cardiogram.

[0007] The mapping system can then be employed to determine the likely source location of an arrhythmia referred to as an inferred source location. To determine the inferred source location for a patient, for each cycle of a patient cardiogram, the mapping system searches the library for a simulated cycle that satisfies a similarity criterion — i.e., is sufficiently similar to the patient cycle.

[0008] Similarity may be measured using a similarity score which may, for example, range from 0.0 to 1 .0 with 1 .0 indicating maximum similarity. A simulatedAttorney Docket No.: 129292.8042. WOOOcycle may be considered similar if its similarity score exceeds a threshold, such as 0.8. The mapping system selects the source location that is mapped to each similar simulated cycle as an inferred source location for the patient’s arrhythmia. A source location score may also be generated based on the similarity score, indicating a level of confidence that the inferred source location corresponds to the actual source location of the arrhythmia.

[0009] To assist an electrophysiologist in planning an ablation, a three-dimensional (3D) graphic of a heart with inferred source locations highlighted may be generated and displayed. The highlighting may be based on the similarity scores, for example, with higher similarity scores indicated by a higher intensity color near a similar inferred source location and lower similarity scores indicated by a lower intensity color near a less similar inferred source location. The highlighting thus reflects a level of confidence that an inferred source location corresponds to the actual source location.

[0010] Although the inferred source locations identified by such a mapping system have significantly improved patient outcomes, they may not precisely match the actual source location of the patient’s arrhythmia. For example, there may be no simulation that is based on simulated cardiac characteristics that precisely match the patient’s cardiac characteristics. In such cases, the source location score may not fully reflect the differences between the simulated cardiac characteristics and the patient’s cardiac characteristics. It would be beneficial to improve patient outcomes to provide a mapping system that would more comprehensively account for such differences.BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 illustrates a 3D graphic with normal tissue, border zone tissue, and scar tissue illustrated in varying shades of gray.

[0012] Figure 2 is a flow diagram that illustrates an SLA method that employs the SLA system to generate a refined source location score for a patient and then performs an ablation procedure.

[0013] Figure 3 is a flow diagram that illustrates the processing of a non-ML refinement component of the SLA assessment system in some embodiments.

[0014] Figure 4 is a flow diagram that illustrates the processing of a train refinement ML model component of the SLA system in some embodiments.Attorney Docket No.: 129292.8042. WOOO

[0015] Figure 5 is a flow diagram that illustrates the processing of a refine initial source location score ML model component of the SLA system in some embodiments.DETAILED DESCRIPTION

[0016] Methods and systems are provided to generate a refined assessment that an inferred source location corresponds to the actual source location of a patient’s arrhythmia. The refinement improves upon an initial assessment, which estimates the likelihood that the inferred source location is the patient’s actual source location. This initial assessment may be based on cardiac characteristics (e.g., cardiac tissue type) that are not patient specific— that is, not derived from the patient’s cardiac characteristics. For example, the initial assessment might assume a typical distribution of cardiac tissue types proximal to the inferred source location which can differ significantly from the patient’s actual distribution. If the patient’s proximal cardiac tissue is all normal or all scar, the inferred source location is unlikely to be the actual origin of the arrhythmia. Conversely, if the proximal cardiac tissue is mostly scar but not entirely, it may indicate a higher likelihood. The refined assessment may be more accurate because it is based on the patient’s cardiac characteristics. These refined assessments support more informed clinical decisions, such as selecting an ablation target that differs from the inferred source location.

[0017] A source location assessment (SLA) system is provided that refines an initial source location score (the initial assessment) of an inferred source location to provide a refined source location score (the refined assessment) that more accurately accounts for the patient’s cardiac characteristics. The SLA system is described primarily in the context of a cardiac characteristic that is cardiac tissue state. However, the SLA system may base the refinement on other cardiac characteristics such as myocardial thickness, myocardial strain, myocardial perfusion, and so on.

[0018] To provide the assessments, the SLA system inputs a patient cardiogram and patient cardiac imaging data that are collected from the patient. The SLA system identifies an inferred source location and generates an initial source location score based on analysis of the patient cardiogram. The SLA system also identifies, based on the patient cardiac imaging data, the cardiac tissue state of cardiac tissue proximal to the inferred source location. The SLA system then generates a refined source location score by adjusting the initial source location score to account for the cardiac tissue stateAttorney Docket No.: 129292.8042. WOOOof the proximal cardiac tissue. Cardiac tissue may be considered proximal if within a certain radius (e.g., 2.0 mm) of the inferred source location. Also, the shape and volume of proximal cardiac tissue may vary based on the inferred source location such as a source location near a pulmonary vein. The proximal cardiac tissue may include myocardial tissue. The inferred source location may be specified as an endocardial location but may be further localized to be a myocardial source location as described in U.S. Pat. No. 12,543,996 titled “Heart Wall Refinement of Arrhythmia Source Locations” and issued on February 10, 2026, which is hereby incorporated by reference.

[0019] In some embodiments, the SLA system applies a mapping system to a patient cardiogram (e.g., one or more cycles (beats) or portions of cycles) of a patient cardiogram to identify an inferred source location and generate an initial source location score. The initial source location score indicates a level of confidence (e.g., a probability) that the inferred source location is the actual source location of the patient’s arrhythmia. Such a mapping system is described in U.S. Pat. No. 10,860,754 titled “Calibration of Simulated Cardiograms” and issued on December s, 2020, which is hereby incorporated by reference. One version of the mapping system may employ a non-ML algorithm (i.e., not based on machine learning (ML)) that accesses a library of mappings that each associates a library cardiogram with a library source location. The mappings may be generated based on simulations of cardiac electrical activity that each assumes a simulated source location of an arrhythmia. In such a case, a library cardiogram is a simulated cardiogram generated based on the simulated cardiac electrical activity, and the associated library source location is the simulated source location. The non-ML algorithm inputs a patient cardiogram, identifies a mapping with a library cardiogram that is similar to the patient cardiogram, and outputs the associated library source location as an inferred source location. The non-ML algorithm also outputs an initial source location score that may be based on the similarity between the patient cardiogram and the library cardiogram. The initial source location score may range from 0.0 to 1 .0 with 0.0 representing not similar and 1 .0 representing very similar. Alternatively, the mapping system may employ an ML model that is trained using mappings and that inputs a patient cardiogram and outputs an inferred source location and an initial source location score.

[0020] In some embodiments, the SLA system inputs patient imaging data that may be a four-dimensional (4D) image of the patient’s heart. The 4D image may beAttorney Docket No.: 129292.8042. WOOOcollected using various imaging techniques such as magnetic resonance (MRI) imaging or computed tomography (CT) imaging. The SLA system analyzes the 4D image to determine various cardiac characteristics such as cardiac tissue state. The patient imaging data may also be a 3D image of the patient’s heart created based on, for example, sestamibi imaging or positron emission imaging. Techniques for determining cardiac tissue state are described in U.S. Pat. No. 11 ,896,432 titled “Machine Learning for Identifying Characteristics of a Reentrant Circuit” and issued on February 13, 2024, which is hereby incorporated by reference. The SLA system generates cardiac tissue scores that indicate the cardiac tissue state of cardiac tissue. The cardiac tissue scores may range from 0.0 to 1.0 with 0.0 representing scar (e.g., dead) tissue, 0.5 representing borderzone tissue, and 1.0 representing normal (e.g., viable) tissue. The SLA may generate a 3D mesh representing the patient’s cardiac geometry (e.g., derived from the patient imaging data). A vertex of the 3D mesh may be designated as corresponding to the inferred source location, and vertices proximal to that vertex are each associated with a cardiac tissue score. Given the initial source location score, the SLA system generates the refined source location score taking into account the patient’s cardiac tissue state (represented by the cardiac tissue scores) of cardiac tissue that is proximal to the inferred source location.

[0021] In some embodiments, the SLA system refines an initial source location score based on an origination score that indicates the suitability of the proximal cardiac tissue for originating and / or sustaining an arrhythmia. The origination score may range from 0.0 to 1 .0 with 0.0 indicating that an arrhythmia is very unlikely to originate and 1 .0 indicating that an arrhythmia is more likely to originate. The SLA system may generate the origination score using various techniques. One technique first determines the average cardiac tissue score of the proximal cardiac tissue. If the proximal cardiac tissue has an equal number of cardiac tissue scores of 0.0, 0.5, and 1 .0, the average cardiac tissue score is 0.5. The origination score may be generated by applying an origination function to the average cardiac tissue score. The origination function returns a maximum origination score of 1 .0 when the average cardiac tissue score has a maximum likelihood of supporting an arrhythmia such as an average cardiac tissue score of 0.75. The origination function may return an origination score of 0.0 when the average cardiac tissue score is 0.0 (all scar) or 1 .0 (all normal). For other average cardiac tissue scores, the origination function may return origination scores thatAttorney Docket No.: 129292.8042. WOOOdecrease linearly based on the difference between that average cardiac tissue score and average cardiac tissue score that has the maximum likelihood of supporting an arrhythmia. For example, if the average cardiac tissue score of 0.75 has the maximum likelihood, the origination score for the average cardiac tissue scores of 0.875 and 0.375 may be 0.5. Thus, if the initial source location score is 0.9 and the origination score is 0.5, the refined source location score would be 0.45 (i.e., 0.9 * 0.5). Other origination functions may be used such as one having a plateau around 0.75 (e.g., 0.65 to 0.80) that decreases linear from the ends of the plateau. Rather than the decrease being linear, the decrease may be exponential, polynomial, logarithmic, sigmoidal, and so on.

[0022] In some embodiments, the SLA system may learn the origination function based on a collection of electronic health records (EHRs) of patients who have had an ablation at an ablation location to treat an arrhythmia. For each EHR, the inferred source location of an arrhythmia and an initial source location score are generated using the mapping system. Also, the cardiac tissue state proximal to the inferred source location is derived from imaging data of the EHR. An average cardiac tissue score is generated based on those cardiac tissue scores. Also, an ablation distance that is the distance between the inferred source location and the ablation location is determined. Thus, for each EHR, the initial source location score, average cardiac tissue score, and ablation distance are known. Curve fitting may be employed to generate a mapping of average cardiac tissue scores to ablation distances. To generate an origination score, the origination function may input an average cardiac tissue score, identify the ablation distance corresponding to that average cardiac tissue score, and output one minus the ratio of that ablation distance and a maximum ablation distance. Thus, the origination score is 1 .0 when the ablation distance is 0.0, and the origination score is 0.0 when the ablation distance is the maximum ablation distance. A refined source location score may be, for example, the origination score times the initial source location score.

[0023] The SLA system may also employ an origination score ML model to generate an origination score. The origination score ML model may be trained using the actual distribution of cardiac tissue state rather than an average cardiac tissue state. The training data may be derived from EHRs. For each EHR, a 3D mesh is generated with vertices corresponding to the proximal cardiac tissue that are each associated with a cardiac tissue state. The training data set includes training examples that each has a feature vector with features derived the cardiac tissue state of the vertices labeledAttorney Docket No.: 129292.8042. WOOOwith an origination score that is one minus the ratio of the ablation distance of that EHR and a maximum ablation distance. Once trained, a 3D mesh for a patient is generated and input to the origination score ML model which outputs an ablation score. This origination score ML model takes into consideration, for example, whether the cardiac tissue state corresponds to a reentrant circuit with an isthmus. Such a reentrant circuit may be more likely to originate an arrhythmia than an average cardiac tissue score would indicate. The origination score ML model also factors in the distance of proximal cardiac tissue from inferred source location. Such a distance may also be employed with a non-ML origination function, for example, by generating a weighted average cardiac tissue score by decreasing the weight of the cardiac tissue scores based on distance of cardiac tissue from the inferred source location.

[0024] In some embodiments, the SLA system may employ origination functions that are specific to an arrhythmia type. The arrhythmia types may include atrial fibrillation, atrial flutter, ventricular tachycardia, premature ventricular contractions, and so on. Origination functions that are specific to arrhythmia type account for differences between the distributions of cardiac tissue type that are more likely to support each arrhythmia type. Each origination function for an arrhythmia type may be derived from EHRs of patients who had an arrhythmia of that type.

[0025] Figure 1 illustrates a 3D graphic with normal tissue 101 , border zone tissue 102, and scar tissue 103 illustrated in varying shades of gray. The source location 104 is illustrated with contours. The SLA system may identify the tissue state (e.g., level of perfusion) from scans of the patient heart. A scan typically generates two-dimensional (2D) images representing slices of a heart. Each 2D image may include colors to indicate the tissue state in the portion of the patient heart represented by the 2D image. The tissue state may alternatively be identified in metadata associated with the 2D images, such as metadata that includes a color or other identifier of the tissue state for each pixel of the image. The metadata may also indicate the portion of the patient heart that the slice represents, such as an orientation and a location on the heart wall. For each 2D image, the SLA system maps that 2D image to a corresponding 3D model slice of a 3D model of a heart representing a heart geometry and including an indication of a source location. The SLA system then adds to that 3D model slice (directly or via metadata) a specification of the tissue state represented by the 2D image. The 3D model may also represent different sublayers of layers of the heart wall. The layers ofAttorney Docket No.: 129292.8042. WOOOa heart include the endocardium, myocardium, and epicardium. The myocardium may have, for example, 16 sublayers spanning the thickness of the myocardium. When a 3D graphic is generated based on the 3D model, the tissue state for a selected sublayer may be displayed. When different sublayers are selected for display (e.g., by a user), an EP may analyze the sublayers to assess the full extent (volume) of the tissue state and source location. In addition, the SLA system may also display a 2D graphic representing a 3D model slice of the 3D model. Such a 2D graphic represents the tissue state and source location across all the sublayers of the heart wall. The SLA system may also adjust the 3D model based on the geometry of the patient heart represented by the 2D images.

[0026] Figure 2 is a flow diagram that illustrates an SLA method that employs the SLA system to generate a refined source location score for a patient and then performs an ablation procedure. The SLA method 200 that performs receiving a patient’s ECG and imaging data, generating a refined source location score using the SLA system, and treating the patient with an ablation procedure factoring in the refined source location score. In block 201 , the method employs a mapping function to determine an inferred source location (ISL) and an initial source location score (ISLS) based on the ECG. In block 202, the method determines the cardiac tissue state (CTS) based on the imaging data. In block 203, the method applies an origination score function to the cardiac tissue state to generate an origination score and multiplies the initial source location score by the origination score to generate a refined source location score (RSLS). In block 204, the method generates a 3D graphic of the patient’s heart based on the imaging data. In block 205, the method adds a hotspot to the 3D graphic to illustrate the inferred source location and indicates the refined source location score. The refined source location score may be displayed as a number or represented by a color intensity level associated with the hotspot. In block 206, the method displays the 3D graphic. In block 207, the method performs an ablation and then completes.

[0027] The computing systems (e.g., network nodes or collections of network nodes) on which the SLA system and the 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, communications links (e.g., Ethernet, Wi-Fi, cellular, and Bluetooth), and so on. The input devices may include keyboards, pointingAttorney Docket No.: 129292.8042. WOOOdevices, touch screens, gesture recognition devices (e.g., for air gestures to rotate a 3D graphic), an augmented reality device (e.g., view the user interface during a medical procedure), head and eye tracking devices, microphones for voice recognition (e.g., to input selection of user interface buttons), and so on. The computing systems may include high-performance computing systems, distributed systems, cloud-based computing systems, client computing systems that interact with cloud-based computing system, desktop computers, laptops, tablets, e-readers, personal digital assistants, smartphones, gaming devices, servers, and so on. The computing systems may access computer-readable media that include computer-readable storage mediums and data transmission mediums. The computer-readable storage mediums are tangible storage means that do not include a transitory, propagating signal. Examples of computer-readable storage mediums include memory such as primary memory, cache memory, and secondary memory (e.g., DVD), and other storage. The computer-readable storage media may have recorded on them or may be encoded with computer-executable instructions or logic that implements the SLA system and the other described systems. The data transmission media are used for transmitting data via transitory, propagating signals or carrier waves (e.g., electromagnetism) via a wired or wireless connection. The computing systems may include a secure crypto processor as part of a central processing unit (e.g., Intel Secure Guard Extension (SGX)) for generating and securely storing keys, for encrypting and decrypting data using the keys, and for securely executing all or some of the computer-executable instructions of the SLA system. Some of the data (e.g., EHRs) sent by and received by the SLA system may be encrypted, for example, to preserve patient privacy (e.g., to comply with government regulations such the European General Data Protection Regulation (GDPR) or the Health Insurance Portability and Accountability Act (HIPAA) of the United States). The SLA system may employ asymmetric encryption (e.g., using private and public keys of the Rivest-Shamir-Adleman (RSA) standard) or symmetric encryption (e.g., using a symmetric key of the Advanced Encryption Standard (AES)).

[0028] The SLA system and the 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. Generally, program modules or components include routines, programs, objects, data structures, and so on that perform tasks or implement data types of the SLA systemAttorney Docket No.: 129292.8042. WOOOand the other described systems. Typically, the functionality of the program modules may be combined or distributed as desired in various examples. Aspects of the SLA system and the other described systems may be implemented in hardware using, for example, an application-specific integrated circuit (ASIC) or a field programmable gate array (FPGA).

[0029] Figure 3 is a flow diagram that illustrates the processing of a non-ML refinement component of the SLA assessment system in some embodiments. The non-ML refinement component 300 generates a refined source location score that is the initial source location score adjusted based on the distribution of proximal cardiac tissue. The component may weight the cardiac tissue state of the cardiac tissue based on distance from the inferred source location. In block 301 , the component selects the next proximal vertex of a 3D mesh representing the inferred source location and the cardiac tissue state of proximal cardiac tissue. In decision block 302, if all the proximal vertices have already been selected, then the component continues at block 305, else the component continues at block 303. In block 303, the component determines the vertex distance between the inferred source location (designated as a vertex) and the selected vertex. In block 304, the component calculates a weighted cardiac tissue score for the selected vertex and then loops to block 301 to select the next proximal vertex. The weighted cardiac tissue score may be the cardiac tissue score multiplied by the ratio of the vertex distance divided by the distance between the source vertex and the farthest away proximal vertex in the direction of the selected vertex. In block 305, the component calculates an average weighted cardiac tissue score that is the sum of the weighted cardiac tissue scores divided by the number of proximal vertices. In block 306, the component applies an origination function to the average weighted cardiac tissue score to generate an origination score for the proximal cardiac tissue. In block 307, the component sets the refined source location score to the initial source location score multiplied by the origination score and completes.

[0030] Figure 4 is a flow diagram that illustrates the processing of a train refinement ML model component of the SLA system in some embodiments. The train refinement ML model component 400 inputs EHRs, generates a training dataset, and trains the ML model to generate weights. In block 401 , the component selects the next EHR. In decision block 402, if all the EHRs have been selected, then the component continues at block 408, else the component continues at block 403. In block 403, theAttorney Docket No.: 129292.8042. WOOOcomponent retrieves the imaging data, ECG, and ablation location from the EHR. In block 404, the component applies the mapping function to the ECG to generate an inferred source location and an initial source location score. In block 405, the component determines the distance between the ablation distance (AL) between the inferred source location and the ablation location. In block 406, the component determines the cardiac tissue state distribution. In block 407, the component generates a training example for the training dataset that includes a feature vector based on the cardiac tissue state distribution and a label based on the ablation distance and then loops to block 401 to select the next EHR. The feature vector may be the imaging data itself (e.g., pixel image) or features derived from the imaging data. In block 408, the component trains the refinement ML model using the training dataset to learn model weights. In block 409, the component stores the weights in a data store and completes.

[0031] Figure 5 is a flow diagram that illustrates the processing of a refine initial source location score ML model component of the SLA system in some embodiments. The refine initial source location score ML model component 500 inputs an initial source location score and a cardiac tissue state distribution and outputs a refined source location score. In block 501 , the component generates a feature vector based on the cardiac tissue score distribution. In block 502, the components retrieves the weights of the refined initial source location score that were learned during the ML model training. In block 503, the component inputs the feature vector to the refine initial source location score ML model which outputs an origination score. In block 504, the component multiplies the initial source location score by the origination score to generate the refined source location score and then completes.

[0032] The ML models employed by the SLA system may be any of a variety or combination of supervised, semi-supervised, self-supervised, unsupervised, or reinforcement learning ML architectures including a neural network such as fully connected, convolutional, recurrent, or autoencoder neural network, transformer, generative adversarial networks, diffusion, and so on. When the ML model is a deep neural network, the model is trained using training data that includes features derived from data and labels corresponding to the data. For example, the data may be an image of proximal cardiac tissue or a cardiac tissue state (determined from imaging data) and a label indicating ablation distance. The training results in a set of weights for the activation functions of the layers of the deep neural network. The trained deep neuralAttorney Docket No.: 129292.8042. WOOOnetwork can then be applied to new data to generate a label for that new data. When the ML model is a support vector machine, a hyper-surface is found to divide the space of possible inputs. An ML model may generate values of discrete domain (e.g., origination scores of 0.0, 0.1 , . . . 1 .0) or values of a continuous domain.

[0033] A neural network model has three major components: architecture, loss function, and search algorithm. The architecture defines the functional form relating the inputs to the outputs (in terms of network topology, unit connectivity, and activation functions). The search for a set of weights that minimizes the loss function is the training process. A neural network model may use a radial basis function (RBF) network and a standard or stochastic gradient descent as the search technique with backpropagation.

[0034] A convolutional neural network (CNN) has multiple layers such as a convolutional layer, a rectified linear unit (ReLU) layer, a pooling layer, a fully connected (FC) layer, and so on. Some more complex CNNs may have multiple convolutional layers, pooling layers, and FC layers. Each layer includes a neuron for each output of the layer. A neuron inputs outputs of prior layers (or original input) and applies an activation function to the inputs to generate an output.

[0035] A convolutional layer may include multiple filters (also referred to as kernels or activation functions). A filter inputs a convolutional window, for example, of an image of an ECG, applies weights to each pixel of the convolutional window, and outputs value for that convolutional window. For example, if the static image is 256 by 256 pixels, the convolutional window may be 8 by 8 pixels. The filter may apply a different weight to each of the 64 pixels in a convolutional window to generate the value. The convolutional layer may include, for each filter, a node (also referred to as a neuron) for each pixel of the image assuming a stride of one with appropriate padding. Each node outputs a feature value based on a set of learned weights for the filter of that node.

[0036] An activation function has a weight for each input and generates an output by combining the inputs based on the weights. The activation function may be an ReLU that sums the values of each input times its weight to generate a weighted value and outputs max(0, weighted value) to ensure that the output is not negative. The weights of the activation functions are learned when training an ML model. The ReLU function of max(0, weighted value) may be represented as a separate ReLU layer with a neuronAttorney Docket No.: 129292.8042. WOOOfor each output of the prior layer that inputs that output and applies the ReLLI function to generate a corresponding “rectified output.”

[0037] A pooling layer may be used to reduce the size of the outputs of the prior layer by downsampling the outputs. For example, each neuron of a pooling layer may input 16 outputs of the prior layer and generate one output resulting in a 16-to-1 reduction in outputs. An FC layer includes neurons that each input all the outputs of the prior layer and generate a weighted combination of those inputs.

[0038] One example of a CNN is a ll-Net ML model. The U-Net ML model includes a contracting path and an expansive path. The contracting path includes a series of max pooling layers to reduce spatial information of the input image and increase feature information. The expansive path includes a series of upsampling layers to convert the feature information to the output image. The input of a U-Net may represent an image such as imaging data corresponding to proximal cardiac tissue, and the output is an origination score.

[0039] The SLA system may also employ a generative adversarial network (GAN) to generate additional training data when, for example, the available training data may not be sufficient for effective training. (See, Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A. and Bengio, Y., 2020. Generative adversarial networks. Communications of the ACM, 63(11), pp.139-144, which is hereby incorporated by reference.) A GAN employs a generator and discriminator and is trained using a training dataset such as that described above for the refinement ML model. The generator generates training data based on random input. The generator is trained to generate training data that cannot be distinguished from real training data. The discriminator indicates whether input training data is real or generated. The generator and discriminator are trained in parallel to learn weights. The generator is trained to generate increasingly more realistic training data, and the discriminator is trained to discriminate between real training data and generated training data more effectively. After being trained, the generator can be used to generate generated training data that is realistic, which can be used in training the refinement ML model.

[0040] The SLA system may employ a diffusion ML model to generate additional training data using a generative process. (See, Rombach, R., Blattmann, A., Lorenz, D., Esser, P. and Ommer, B., 2022. High-resolution image synthesis with latent diffusionAttorney Docket No.: 129292.8042. WOOOmodels. In Proceedings of the IEEE / CVF conference on computer vision and pattern recognition (pp. 10684-10695), which is hereby incorporated by reference.) A diffusion ML model is a generative ML model that inputs noisy data and progressively denoises the data until the denoised data appears to be indistinguishable from real data such as imaging data of proximal cardiac tissue. A diffusion ML model is trained using a forward diffusion process that successively adds noise to input training data such as imaging data to generate noisy data and a reverse diffusion process that successively denoises the noisy data to generate denoised data that approximates the input training data. The training learns weights for the reverse diffusion process that tend to minimize the difference between the input training data and the denoised data. After a diffusion model is trained, the reverse diffusion process is employed to generate data that can be used to train the refinement ML model. To do so, randomly generated noisy data is input to the reverse diffusion process which denoises the noisy data to generate the noised data that appears to be real data.

[0041] The forward diffusion process employs a Markov chain that incrementally adds Gaussian noise to the training data over a series of steps. This process transforms the training data from its initial distribution to a Gaussian distribution. The reverse diffusion process employs a neural network to incrementally approximate and remove the noise that was added at each step of the forward diffusion process. When generating data, randomly generated noisy data is input to the reverse diffusion process which incrementally removes the noise that was learned during training.

[0042] The forward diffusion process systematically adds Gaussian noise to the original data xo Gaussian noise over T timesteps, resulting in a sequence of increasingly noisy data xi, X2, . . . , ,XT. The process at each time step t may be represented by the equation:

[0043] where xtis data at timestep t, etis Gaussian noise, atis the amount of noise added, and I is the identity matrix.

[0044] The reverse diffusion process learns the distribution of the training data by starting from noise and progressively denoising it over the timesteps. The trainingAttorney Docket No.: 129292.8042. WOOOestimates the reverse of the forward diffusion process using a neural network that may be represented by the equation:«<-< = .- H' a-

[0045] where t:represent the cumulative noise and ’ represents the neural network.

[0046] The goal of training a diffusion model is to minimize the difference between the original data and the data reconstructed by the reverse diffusion process using a loss function that may be represented by the equation.

[0047] In some embodiments, the SLA system may employ a K-Nearest Neighbor (KNN) model to provide a refined source location score. The training data for a KNN model may be a training dataset with training examples that each have a feature vector with features that may be derived from imaging data such as cardiac tissue state. Each also has a label for that feature vector indicating information relating to a patient (e.g., origination score) associated with those features. A KNN model may be used without a training phase that is without learning weights or other parameters to represent the training data. In such a case, the patient feature vector is compared to the training feature vectors to identify a number (e.g., represented by the “K” in KNN) of similar training feature vectors. Once the number of similar training feature vectors are identified, the labels associated with the similar training feature vectors are analyzed to provide information (e.g., origination score) for the patient. The labels of the training feature vectors that are more similar to a patient feature vector may be given a higher weight than those that are less similar. For example, if k is 10 and four training feature vectors are very similar and six are less similar, similarity weights of 0.9 may be assigned to the very similar training feature vectors and 0.2 to the less similar. If three of the four and one of the six have the same information, then the information for the patient is primarily based on that information even though most of the 10 have different information. Conceptually, training feature vectors that are very similar are closer to the patient’s feature vector in a multi-dimensional space of features and a similarity weight is based on distance between the feature vectors. Various techniques may beAttorney Docket No.: 129292.8042. WOOOemployed to calculate a similarity metric indicating similarity between a patient feature vector and a training feature vector such as a dot product, cosine similarity, Pearson correlation, and so on.

[0048] The following paragraphs describe various aspects of the SLA system. An implementation of the SLA system may employ any combination or sub-combination of the aspects and may employ additional aspects. The processing of the aspects may be performed by one or more computing systems with one or more processors that execute computer-executable instructions that implement the aspects and that are stored on one or more computer-readable storage mediums.

[0049] In some aspects, the techniques described herein relate to a method for treating a patient having an arrhythmia, the method including: performing under control of one or more computing systems the following: accessing patient data that includes a patient cardiogram and patient cardiac imaging data; applying a mapping system to the patient cardiogram to identify an inferred source location of the arrhythmia and an initial source location score indicating confidence in the inferred source location being an actual source location of the arrhythmia; analyzing the patient cardiac imaging data to identify cardiac tissue state of cardiac tissue; generating cardiac tissue state scores that indicate cardiac tissue state of the cardiac tissue; generating a refined source location score based on the initial source location score and the cardiac tissue state scores, the refined source location score indicating a refined confidence in the inferred source location being the actual source location of the arrhythmia factoring in the cardiac tissue state; and displaying a graphic of a heart that indicates the inferred source location and an indication of refined source location score; and performing an ablation on a patient is based at a source location that is determined factoring in the inferred source location and the refined source location score. In some aspects, the techniques described herein relate to a method wherein the generating of the refined source location score is based on an origination score for proximal cardiac tissue that indicates suitability of the proximal cardiac tissue in originating and / or sustaining an arrhythmia. In some aspects, the techniques described herein relate to a method wherein the origination score is based on similarity of a distribution of the cardiac tissue state of the patient to distributions whose suitability to originate and / or sustain an arrhythmia is known. In some aspects, the techniques described herein relate to a method wherein the similarity based on applying a machine learning model to patient cardiac tissue state to generateAttorney Docket No.: 129292.8042. WOOOan indication the suitability. In some aspects, the techniques described herein relate to a method wherein the suitability is indicated by the origination score. In some aspects, the techniques described herein relate to a method wherein the cardiac tissue includes myocardial tissue. In some aspects, the techniques described herein relate to a method wherein the generating of the cardiac tissue state scores is based on perfusion, myocardial thickness, and / or myocardial strain represented by patient cardiac imaging data. In some aspects, the techniques described herein relate to a method wherein cardiac tissue states scores are assigned to vertices of a 3D mesh, the vertices corresponding to locations of proximal cardiac tissue.

[0050] In some aspects, the techniques described herein relate to one or more computing systems for refining an initial source location score of an arrhythmia of a patient, the one or more computing systems including: one or more computer-readable storage mediums that store computer-executable instructions for controlling the one or more computing systems to: access a patient cardiogram and patient cardiac imaging data; apply a mapping system to the patient cardiogram to identify an inferred source location of the arrhythmia and an initial source location score indicating confidence in the inferred source location being an actual source location of the arrhythmia; analyze the patient cardiac imaging data to identify cardiac tissue state of cardiac tissue; generate the refined source location score based on the initial source location score and the cardiac tissue state, the refined source location score indicating a refined confidence in the inferred source location being the actual source location of the arrhythmia; and display a graphic of a heart that indicates the inferred source location and an indication of refined source location score; and one or more processors for controlling the one or more computing systems to execute one or more of the computerexecutable instructions.

[0051] In some aspects, the techniques described herein relate to a system for refining an assessment of a source location being an actual source location of an arrhythmia, including: a processor configured to: identify an inferred source location of the arrhythmia; generate an initial source location score representing a likelihood that the inferred source location is an actual source of the arrhythmia; determine cardiac tissue state of cardiac tissue proximal to the inferred source location; and generate a refined source location score by adjusting the initial source location score based on the cardiac tissue state of the cardiac tissue proximal to the inferred source location.Attorney Docket No.: 129292.8042. WOOO

[0052] In some aspects, the techniques described herein relate to a method for treating a patient based on a refinement of a source location assessment of an arrhythmia, including: receiving a patient cardiogram; identifying an inferred source location of an arrhythmia based on the patient cardiogram; generating an initial source location score representing a likelihood that the inferred source location is an actual source of the arrhythmia; receiving patient cardiac imaging data; determining a cardiac tissue state of cardiac tissue that is proximal to the inferred source location from the patient cardiac imaging data; calculating a refined source location score by modifying the initial source location score based on the cardiac tissue state; and performing an ablation on the patient targeting a location that is identified factoring in the inferred source location and the refined source location score. In some aspects, the techniques described herein relate to a method, wherein the cardiac tissue state includes a cardiac tissue score ranging from 0.0 to 1 .0, with 0.0 representing scar tissue and 1 .0 representing normal tissue. In some aspects, the techniques described herein relate to a method, wherein the proximal cardiac tissue includes tissue within a predefined radius of the inferred source location. In some aspects, the techniques described herein relate to a method, wherein the refined source location score is calculated by multiplying the initial source location score by an origination score derived from the cardiac tissue state. In some aspects, the techniques described herein relate to a method, wherein the origination score is generated using an origination function that returns a maximum likelihood when the average cardiac tissue score is within a predetermined range. In some aspects, the techniques described herein relate to a method, wherein the origination function is learned from electronic health records including historical ablation outcomes. In some aspects, the techniques described herein relate to a method, wherein the origination score is produced using a machine learning model trained on 3D meshes representing cardiac tissue states. In some aspects, the techniques described herein relate to a method, wherein the refined source location score is further based on a type of arrhythmia selected from the group consisting of atrial fibrillation, atrial flutter, ventricular tachycardia, and premature ventricular contractions. In some aspects, the techniques described herein relate to a method, further including generating a 3D mesh from the patient cardiac imaging data, wherein vertices of the mesh are associated with cardiac tissue scores. In some aspects, the techniquesAttorney Docket No.: 129292.8042. WO00described herein relate to a method, wherein the cardiac imaging data includes 4D magnetic resonance imaging or computed tomography imaging.

[0053] All documents incorporated by reference are incorporated in their entirety for the full extent of their disclosures. In the event of inconsistencies between the language in this document and any incorporated-by-reference document, the language in the incorporated-by-reference document should be considered supplementary to that of this document and the language in this document controls.

[0054] Although the subject matter has been described in language specific to 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.

Claims

Attorney Docket No.: 129292.8042. WOOOCLAIMSl / We claim:

1. A method for treating a patient having an arrhythmia, the method comprising:performing under control of one or more computing systems the following: accessing patient data that includes a patient cardiogram and patient cardiac imaging data;applying a mapping system to the patient cardiogram to identify an inferred source location of the arrhythmia and an initial source location score indicating confidence in the inferred source location being an actual source location of the arrhythmia;analyzing the patient cardiac imaging data to identify cardiac tissue state of cardiac tissue;generating cardiac tissue state scores that indicate cardiac tissue state of the cardiac tissue;generating a refined source location score based on the initial source location score and the cardiac tissue state scores, the refined source location score indicating a refined confidence in the inferred source location being the actual source location of the arrhythmia factoring in the cardiac tissue state; anddisplaying a graphic of a heart that indicates the inferred source location and an indication of refined source location score; and performing an ablation on a patient based at a source location that is determined factoring in the inferred source location and the refined source location score.

2. The method of claim 1 wherein the generating of the refined source location score is based on an origination score for proximal cardiac tissue that indicates suitability of the proximal cardiac tissue in originating and / or sustaining an arrhythmia.Attorney Docket No.: 129292.8042. WOOO3. The method of claim 2 wherein the origination score is based on similarity of a distribution of the cardiac tissue state of the patient to distributions whose suitability to originate and / or sustain an arrhythmia is known.

4. The method of claim 3 wherein the similarity based on applying a machine learning model to patient cardiac tissue state to generate an indication the suitability.

5. The method of claim 4 wherein the suitability is indicated by the origination score.

6. The method of claim 1 wherein the cardiac tissue includes myocardial tissue.

7. The method of claim 1 wherein the generating of the cardiac tissue state scores is based on perfusion, myocardial thickness, and / or myocardial strain represented by the patient cardiac imaging data.

8. The method of claim 1 wherein cardiac tissue states scores are assigned to vertices of a three-dimensional (3D) mesh, the vertices corresponding to locations of proximal cardiac tissue.

9. One or more computing systems for refining an initial source location score of an arrhythmia of a patient, the one or more computing systems comprising:one or more computer-readable storage mediums that store computerexecutable instructions for controlling the one or more computing systems to:access a patient cardiogram and patient cardiac imaging data; apply a mapping system to the patient cardiogram to identify an inferred source location of the arrhythmia and an initial source location score indicating confidence in the inferred source location being an actual source location of the arrhythmia;analyze the patient cardiac imaging data to identify cardiac tissue state of cardiac tissue;Attorney Docket No.: 129292.8042. WOOOgenerate the refined source location score based on the initial source location score and the cardiac tissue state, the refined source location score indicating a refined confidence in the inferred source location being the actual source location of the arrhythmia; and display a graphic of a heart that indicates the inferred source location and an indication of refined source location score; andone or more processors for controlling the one or more computing systems to execute one or more of the computer-executable instructions.

10. A system for refining an assessment of a source location being an actual source location of an arrhythmia, comprising:a processor configured to:identify an inferred source location of the arrhythmia;generate an initial source location score representing a likelihood that the inferred source location is an actual source of the arrhythmia; determine cardiac tissue state of cardiac tissue proximal to the inferred source location; andgenerate a refined source location score by adjusting the initial source location score based on the cardiac tissue state of the cardiac tissue proximal to the inferred source location.

11. A method for treating a patient based on a refinement of a source location assessment of an arrhythmia, comprising:receiving a patient cardiogram;identifying an inferred source location of an arrhythmia based on the patient cardiogram;generating an initial source location score representing a likelihood that the inferred source location is an actual source of the arrhythmia; receiving patient cardiac imaging data;determining a cardiac tissue state of cardiac tissue that is proximal to the inferred source location from the patient cardiac imaging data;calculating a refined source location score by modifying the initial source location score based on the cardiac tissue state; andAttorney Docket No.: 129292.8042. WOOOperforming an ablation on the patient targeting a location that is identified factoring in the inferred source location and the refined source location score.

12. The method of claim 11 , wherein the cardiac tissue state includes a cardiac tissue score ranging from 0.0 to 1 .0, with 0.0 representing scar tissue and 1 .0 representing normal tissue.

13. The method of claim 11, wherein the proximal cardiac tissue includes tissue within a predefined radius of the inferred source location.

14. The method of claim 11 , wherein the refined source location score is calculated by multiplying the initial source location score by an origination score derived from the cardiac tissue state.

15. The method of claim 14, wherein the origination score is generated using an origination function that returns a maximum likelihood when the average cardiac tissue score is within a predetermined range.

16. The method of claim 15, wherein the origination function is learned from electronic health records comprising historical ablation outcomes.

17. The method of claim 14, wherein the origination score is produced using a machine learning model trained on three-dimensional meshes representing cardiac tissue states.

18. The method of claim 11 , wherein the refined source location score is further based on a type of arrhythmia selected from the group consisting of atrial fibrillation, atrial flutter, ventricular tachycardia, and premature ventricular contractions.

19. The method of claim 11, further comprising generating a three-dimensional mesh from the patient cardiac imaging data, wherein vertices of the mesh are associated with cardiac tissue scores.Attorney Docket No.: 129292.8042. WOOO20. The method of claim 11 , wherein the cardiac imaging data comprises fourdimensional magnetic resonance imaging or computed tomography imaging.

Images