System and method for non-invasive localization of right ventricular activation sites

Abstract

A system includes a processor and a storage medium having encoded thereon executable instructions that, when executed by the processor, cause the processor to carry out a method. The method includes receiving surface electrocardiogram (ECG) data of a patient, translating the surface ECG data into 3D spatial coordinates on a right ventricle (RV) anatomical model, and predicting an origin of a ventricular arrhythmia (VA), based on the translated ECG data, on the RV anatomical model.

Description

Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026SYSTEM AND METHOD FOR NON-INVASIVE LOCALIZATION OF RIGHT VENTRICULAR ACTIVATION SITESRELATED APPLICATION(S)

[0001] This application claims the benefit of and priority to U.S. Provisional Application No. 63 / 790,605. filed April 17, 2025. the contents of which is incorporated herein by reference.GOVERNMENT SUPPORT

[0002] This invention was made with government support under GrantR15HL 167222 which was awarded by the National Institute of Health. The Government has certain rights in the invention.BACKGROUND

[0003] Sudden cardiac arrest is a leading cause of mortality in developed countries, accounting for approximately 350,000 deaths per year in the United States. The majority of these are caused by ventricular arrhythmia, most commonly premature ventricular contractions and ventricular tachycardia.

[0004] Catheter ablation is a cornerstone therapy for ventricular arrhythmias, yet accurately localizing arrhythmia origins — particularly in the right ventricle (RV) — still relies heavily on time-consuming and invasive procedures. There is a growing need for non-invasive, rapid, and cost-effective localization tools that can assist or even replace early-stage mapping, especially in settings lacking advanced EAM capabilities.SUMMARY

[0005] The present disclosure relates to a system, including: at least one processor; and at least one storage medium having encoded thereon executable instructions that, when executed by the at least one processor, cause the at least one processor to carry out a method including: receiving surface electrocardiogram (ECG) data of a patient; translating the surface ECG data into 3D spatial coordinates on a right ventricle (RV) anatomical model; and predicting an origin of a ventricular arrhythmia (VA), based on the translated ECG data, on the RV anatomical model.Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026

[0006] The present disclosure relates to a method, including: receiving surface electrocardiogram (ECG) data of a patient; translating the surface ECG data into 3D spatial coordinates on a right ventricle (RV) anatomical model; and predicting an origin of a ventricular arrhythmia (VA), based on the translated ECG data, on the RV anatomical model.

[0007] The present disclosure relates to at least one computer-readable storage medium encoded with computer-executable instructions that, when executed by at least one processor, cause the at least one processor to carry out a method including: receiving surface electrocardiogram (ECG) data of a patient; translating the surface ECG data into 3D spatial coordinates on a right ventricle (RV) anatomical model; and predicting an origin of a ventricular arrhythmia (VA), based on the translated ECG data, on the RV anatomical model.BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The present disclosure is further described in the detailed description which follows, in reference to the noted plurality of drawings by way of non-limiting examples of exemplary embodiments, in which like reference numerals represent similar parts throughout the several views of the drawings, and wherein:

[0009] FIG. 1 A illustrates an embodiment of a system for translating ECG morphology into 3D spatial coordinates on a RV anatomical model, according to one or more embodiments herein;

[0010] FIG. IB illustrates an example method that can be carried out by the system of FIG. 1A, according to one or more embodiments herein;

[0011] FIG. 2 illustrates a generic right ventricular endocardial surface represented in 3D space;

[0012] FIG. 3 illustrates localization error in mm of arrhythmia origin predictions in the RV-endocardial mesh based on 1000-trial bootstrapped RBF SVR and linear SVR models for all JQRS interval windows (30 ms, 40 ms, .... 160 ms) with SD bars, according to one or more embodiments herein;

[0013] FIG. 4 illustrates localization error in mm measured as Euclidean distance of pacing-site origin prediction on the RV-endocardial surface for design set (left panel, n=182) and test set (right panel, n=45) for 60 ms JQRS Interval for RBF and linear methods, according to one or more embodiments herein;

[0014] FIG. 5 A illustrates a recorded, unknown pacing-site ECG, according to one or more embodiments herein;Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026

[0015] FIG. 5B illustrates a predicted pacing site location over an RV mesh, according to one or more embodiments;

[0016] FIG. 5C illustrates the predicted pacing site location on a centroid of a triangular element of a generic RV surface mesh;

[0017] FIG. 6 depicts an exemplary computing device, according to one or more embodiments herein; and

[0018] FIG. 7 depicts a heatmap of significance in localization accuracy between QRS interval windows, according to one or more embodiments herein.

[0019] While the above-identified drawings set forth presently disclosed embodiments, other embodiments are also contemplated, as noted in the discussion. This disclosure presents illustrative embodiments by way of representation and not limitation. Numerous other modifications and embodiments can be devised by those skilled in the art which fall within the scope and spirit of the principles of the presently disclosed embodiments.DETAILED DESCRIPTION

[0020] The present disclosure provides a non-invasive, ECG-based systems and methods that use machine learning to precisely localize right ventricular arrhythmia origins, reducing reliance on invasive mapping during catheter ablation procedures. The systems and methods of the present disclosure can be used for non-invasively localizing the origin of right ventricular (RV) arrhythmias using standard surface ECG data. This can enable physicians to pre-identify likely activation sites and guide ablation without requiring full electroanatomical mapping. In some embodiments, the systems and methods of the present disclosure utilize a machine-leaming-driven framework that can translate multi-lead ECG morphology' into 3D spatial coordinates on a generic RV anatomical model. By combining high-resolution ECG analysis with anatomically grounded machine learning, the systems and methods of the present disclosure provide a physical (computational) solution to the problem of arrhythmia localization — without requiring invasive mapping catheters or costly 3D mapping systems.

[0021] While many cases of ventricular arrhythmia (VA) originate in the left ventricle (LV), idiopathic right-sided VAs have also been shown to trigger sudden cardiac arrest in otherwise healthy individuals. This highlights the need for accurate localization algorithms or models for both LV- and RV-origin VAs. Such tools can optimize pre-procedural planning, improve ablation efficiency, and reduce VA recurrence. In particular, VAs originating from the RV outflow tract (RVOT) and in patients with structural heart disease (e.g., arrhythmogenicAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 right ventricular cardiomyopathy, ARVC) may arise from the RVOT, lateral wall, inferior wall, or apex regions, underscoring the clinical significance of these RV regions.

[0022] Linear regression (MLR) methods to localize VA origins on a generic LV endocardial surface have been used. However, linear mapping of RV presents unique challenges due to the RV’s distinct anatomical features and significantly thinner myocardial wall. The RV’s reduced muscle mass may minimize transmural conduction delays between the epicardium and endocardium, potentially shortening the QRS duration required to capture full RV depolarization. These anatomical and electrophysiological differences may affect both the optimal QRS duration required for full RV depolarization and the validity of linear mapping between pacing locations and ECG morphology features. The present disclosure determined the optimal QRS interval for RV-based localization and extended MLR-based modeling to RV context.

[0023] In some embodiments, machine learning-based localization frameworks are used to predict VA origins on a generic RV mesh surface. Specifically, the methods and systems of the present disclosure (1) identify the optimal QRS interval for accurately capturing RV depolarization, considering its unique anatomical characteristics, and (2) develop and validate a machine-learning model using RV pacing data to localize VA origins on the generic RV surface. Establishing a robust RV-specific localization framework may significantly improve pre-procedural planning, enhance ablation outcomes, and reduce the risk of VA recurrence.

[0024] In reference to FIG. 1A. a system 1 for carrying out methods of the present disclosure is depicted. The system 1, in some embodiments, can be a system for a machine-1 earning-driven framework that translates multi-lead ECG morphology into 3D spatial coordinates on a generic RV anatomical model. In some embodiments, the system 1, includes the following components:

[0025] In some embodiments, the system 1 includes an input feature extraction functional facility 10. In some embodiments, the input feature extraction functional facility 1 is configured to receive data from a data acquisition device 4 regarding a patient 2. For example, a 12-lead electrocardiogram (ECG) can be processed to compute QRS integrals (JQRS ) from eight independent leads (I, II, VI -V6). These values can optionally be trimmed to identify time-restricted segments of interest (e.g., 30-160 ms), improving predictive precision. In some embodiments, a good segment of interest is one that captures the dominant portion of ventricular depolarization while minimizing the inclusion of noise, baseline drift, or irrelevantAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 post-depolarization components. It can be identified based on either (1) a fixed time window relative to the pacing stimulus or QRS onset (e.g., 30-160 ms), or (2) data-driven optimization, where various time intervals are tested to determine which yields the lowest prediction error during model training and validation. The segment may also be trimmed dynamically to exclude periods of low signal amplitude or poor signal-to-noise ratio using thresholds derived from the ECG waveform or prior clinical knowledge. In some embodiments, the recorded 12-lead ECG signals are processed to extract quantitative features such as QRS integrals or timedomain waveform characteristics. These features serve as spatial indicators of electrical activation patterns and form the input for subsequent computational analysis, which is discussed further below. In some embodiments, the QRS integrals, which can optionally be trimmed (e.g., as discussed above), serve as spatial indicators of electrical activation and form the input for subsequent computational analysis, which is discussed further below. In some embodiments, the input feature extraction functional facility 10 can receive a combination of surface electrocardiogram (ECG) recordings and medical imaging data (e.g., CT or MRI) collected from the patient using clinical-grade data acquisition devices. The ECG data captures the temporal characteristics of cardiac electrical activity, while imaging provides anatomical detail for geometric reconstruction.

[0026] In some embodiments, the system includes an anatomical model integration functional facility 20. Generally, the anatomical model integration functional facility 20 can provide a geometric framework onto which patient electrical features (e.g., ECG, QRS data, or trimmed QRS data) are mapped. In some embodiments, a pre-constructed 3D RV mesh model derived from CT imaging (or from other imaging modality) can be used as the anatomical reference frame. In some embodiments, the model can be discretized into segments, with each segment discretized into individual elements, with centroids of the elements representing a candidate pacing or arrhythmia origin site. In some embodiments, the model can include one or more of the following anatomical segments: 1. Anterior RVOT. 2. RVOT left septal, 3. RVOT right septal, 4. Posterior RVOT, 5. Lateral RVOT, 6. Apical septal, 7. Apical inferior, 8. Apical anterior, 9. Middle anterior septal, 10. Middle inferior septal, 11. Middle inferior, 12. Middle lateral, 13. Middle anterior, 14. Basal anterior septal, 15. Basal inferior septal 16. Basal inferior, 17. Basal lateral, or 18. Basal anterior. In some embodiments, the model can include all 18 segments. In some embodiments, the surface of the RV model is discretized into 277 triangular elements, with each centroid representing a candidate pacing or arrhythmia origin site. The RV surface of the model can be discretized into any number of triangular elementsAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 depending on the size of the CT (or other modality) imaging data set used to generate the 3D RV mesh model. In some cases, the larger the imaging data set, the more triangular elements the surface of the RV mesh model can be discretized into, which in turn can increase the accuracy of VA localization on the model. However, it should be appreciated that the mesh model of the RV can be discretized into any number of segments and discrete elements, and the segments and elements can be of any desirable shape. In some embodiments, the segments may be anatomical or non-anatomical segments. In some embodiments, the mesh model can be discretized into between 10 and 30 segments. In some embodiments, the mesh model can be discretized into between 15 and 20 segments. In some embodiments, the mesh model can be discretized into between 200 and 400 discrete elements. In some embodiments, the mesh model can be discretized into between 200 and 400 discrete elements. In some embodiments, the mesh model can be discretized into between 250 and 350 discrete elements. In some embodiments, the mesh model can be discretized into between 250 and 300 discrete elements. In some embodiments, the pre-constructed 3D RV mesh model can be patient generic. For instance, the 3D RV mesh model can be constructed from imaging data generated from a plurality of patients.

[0027] In some embodiments, a patient specific RV mesh model can be prepared by fitting 3D imaging data gathered of the patient 2 (e.g., of the patient’s heart) to the generic, pre-constructed 3D RV mesh model (e.g., the discretized RV surface of the model). For instance, in some embodiments, the anatomical integration functional facility' 20 can receive medical imaging data, from a data acquisition device 4, of the patient 2. In some embodiments the anatomical integration functional facility 20 can receive medical imaging data (e.g., CT or MRI) collected from the patient 2 using cl ini cal -grade data acquisition devices. The imaging provides anatomical detail for geometric reconstruction. The patient specific model can be prepared to enhance anatomical fidelity and improve prediction accuracy. This is accomplished by processing the patient’s individual CT or cardiac magnetic resonance imaging (MRI) data. In some embodiments, the generic, pre-constructed model can be mapped to the patient's 3D model (e.g., the data gathered by the acquisition device 4 of the patient 2) using a rigid registration method. In other embodiments, instead of fitting the patient's cardiac imaging data to the generic, pre-constructed 3D RV mesh model, the preconstructed 3D RV mesh model can be patient-specific, itself. For instance, the 3D RV mesh model can be constructed from imaging data generated solely7from the patient 102.Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026

[0028] In some embodiments, to fit, or map, the patient-specific imaging data to the generic RV model, patient-specific cardiac imaging data, e.g. CT or MRI data, may first be segmented to reconstruct the RV surface of the patient. Following this, the generic RV mesh may then be aligned to the patient-specific geometry using rigid registration and. in some examples, refined using landmark-based, surface-based, or deformable registration. In some embodiments, after alignment, predicted coordinates on the generic RV mesh are projected onto the corresponding patient-specific RV surface location, for example by nearest-surface-point mapping, nearest-centroid mapping, or interpolation over the aligned mesh. It should be appreciated that while rigid registration is discussed throughout the disclosure for fitting, or mapping the patient-specific imaging data to the generic RV model (or vice-versa), that this is merely an illustrative example, and any number of other registration techniques may be used.

[0029] In some embodiments, the integration step can ensure spatial consistency between electrical signals and the patient-specific cardiac structure. In some embodiments, the anatomical model integration functional facility 20 can use cardiac imaging data, such as contrast-enhanced CT or MRI, for anatomical reconstruction and scar identification. The anatomical reconstruction workflow performed by the anatomical model integration functional facility 20 can include image segmentation, surface mesh generation, and registration and projection. The image segmentation can be of the cardiac chambers, particularly the RV, using thresholding or deep learning-based tool. During surface mesh generation, the segmented volume is converted into a triangulated surface representation using algorithms such as marching cubes or mesh simplification techniques. During registration and projection, the generic model can optionally be registered onto patientspecific geometry (e.g., the imaging data generated of the patient 2) to align spatial coordinate systems, enabling feature transfer or model fusion.

[0030] In some embodiments, the system includes a machine learning engine functional facility 30. A trained Support Vector Regression (SVR) model can map ECG features (e.g., the ECG data, such as QRS integrals, gathered of the patient 2) to spatial coordinates (x, y, z) on the RV mesh (e.g., the generic pre-constructed mesh model or the patient's 3D anatomical model mapped to the generic pre-constructed mesh model), for instance as part of the prediction output functional facility 40. While an SVR model is discussed in detail herein, it should be appreciated that this is merely one example of a model that can be incorporated in the present disclosure. Both linear and Gaussian (RBF) kernelAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 versions are included. The model can be optimized using bootstrapped cross-validation to identify the most informative QRS interval and minimize localization error. In some embodiments, a supervised learning model, trained on labeled pacing data or simulated arrhythmia events, as part of the prediction output functional facility 40 for instance, processes the extracted features to estimate the spatial location of the arrhythmic source. In some embodiments, the model leverages both electrical and anatomical information to infer the likely origin or circuit of the ventricular arrhythmia. This framework enables real-time, data-driven estimation of arrhythmic origin sites, supporting clinicians during catheter ablation procedures or diagnostic evaluation. In some embodiments, the input to the SVR model consists of pre-processed ECG features, specifically QRS integrals ( / QRS) computed from eight independent leads (I, II, and V1-V6). Each QRS integral can be calculated over a designated time window (e.g., 30-160 ms), and multiple such windows may be evaluated to determine the optimal temporal segment for localization. The input vector for each sample is an 8-dimensional numerical array: Input = [fQRSJ, jQRSJI, jQRS Vl, JQRS_V2, JQRS V3, JQRS V4, JQRS_V5, JQRS_V6]. These values can be standardized (e.g., z-score normalized) to ensure consistent feature scaling before being passed to the regression model.

[0031] The SVR model can be trained on a dataset of known pacing sites, where each input vector (QRS integrals) is associated with a known ground truth coordinate (x, y, z) on the RV surface. The training procedure can include data partitioning, hyperparameter tuning, model fitting, and performance evaluation. In data partitioning, the dataset can be split into training and validation subsets using bootstrapped cross-validation. In hyperparameter tuning, grid search can be applied over the kernel parameters (e.g., C. epsilon, and gamma for RBF) to minimize prediction error. In model fitting, the SVR model can learn the regression function mapping ECG features to spatial coordinates. In performance evaluation, localization accuracy can be quantified using Euclidean distance error between predicted and true site locations across validation folds.

[0032] More specifically with respect to training the model, in some embodiments, each training sample is an 8-dimensional feature vector built from QRS integrals of leads I, II, and V1-V6. Each training sample can be labeled with a known ground-truth RV surface coordinate corresponding to a pacing-site location or triangle centroid on the RV mesh. In some embodiments, separate regression models may be trained for x, y, and z coordinates. In some embodiments, features may be standardized before training. In some embodiments, hyperparameters may be selected by grid search or random search. In some embodiments,Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 candidate QRS windows may be compared using cross-validation and / or bootstrapping. In some embodiments, the predicted coordinate may be snapped or mapped to the nearest mesh centroid to preserve anatomical consistency.

[0033] Once trained, the model, for instance as part of the prediction output functional facility 40, can infer the (x, y, z) coordinates of an unknown arrhythmia site from ECG data obtained during clinical procedures (e.g., the ECG data, such as QRS integrals, gathered of the patient 2). For instance, the system can extract quantitative ECG features, such as full or trimmed QRS integrals, from the patient ECG data. In some embodiments, the system can normalizes those extracted features. The resulting vector can be input into a trained model. The model then outputs a continuous estimated spatial location, such as x, y, and z coordinates in the coordinate system of the generic RV mesh. That estimated location can then be mapped to a discrete anatomical site on the generic RV mesh, such as a nearest triangle centroid or neighboring candidate region. In some embodiments, the estimated location can be projected to patient-specific geometry if patient-specific imaging data is available.

[0034] The predicted spatial location is mapped to the corresponding triangle centroid on the RV mesh (e.g., the generic pre-constructed mesh model or the patient’s 3D anatomical model mapped to the generic pre-constructed mesh model). This prediction is visualized in 3D on the anatomical model via the software interface (see below software implementation functional facility 50), reported as a coordinate output, and optionally exported for integration with navigation systems or electroanatomical mapping tools.

[0035] In some embodiments, the system includes a prediction output functional facility 40. Given an input ECG (e.g., the ECG data, such as QRS integrals, gathered of the patient 2), the system can predict a likely origin site on the RV mesh, which can be visualized in a custom software interface (e.g., the generic pre-constructed mesh model or the patient’s 3D anatomical model mapped to the generic pre-constructed mesh model). Specifically, the output of the machine learning model is a predicted site or region of interest on the cardiac geometry, indicating the estimated exit site or propagation pathway of the arrhythmia, and this output can be visualized or quantitatively assessed for diagnostic support. Localization error is reported using Euclidean distance between predicted and known sites (when available).

[0036] The prediction process can include the steps, feature extraction, model inference, and anatomical mapping, which can be sequential. In feature extraction, the 12-lead ECG data from the patient 2 is processed by the input feature extraction functional facility to compute QRS integrals from eight independent leads (I, II, and V1-V6). A fixed 60 msAttomey Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 segment starting at the onset of the QRS complex can be selected as the feature window to capture the critical ventricular depolarization waveform. In model inference, the extracted features can then be passed to the trained model, such as the SVR model, within the machine learning engine functional facility. The model can infer a predicted three-dimensional spatial coordinate (x, y, z) corresponding to the likely site of origin. In anatomical mapping, the predicted coordinate can be mapped to the closest centroid on a generic right ventricular (RV) surface mesh comprising elements. If patient-specific RV geometry is available (e.g., reconstructed from CT or MRI). the predicted site is automatically projected from the generic model onto the corresponding location on the personalized RV surface. This ensures spatial alignment with the patient’s anatomical structure, enhancing clinical relevance.

[0037] In some embodiments, the prediction output functional facility' 40 provides a user-facing graphical interface integrated into a custom software platform. The interface can include a 3D visualization panel, which displays the RV mesh with highlighted regions showing the predicted arrhythmia site. The mesh can be rotated, zoomed, and color-coded to reflect prediction confidence or error distance. The interface can include a coordinate output display, which shows the (x, y, z) coordinate of the predicted site numerically . The interface can include overlay features, which can overlay ground truth sites, error vectors, or scar regions from cardiac imaging, if available. The interface can include prediction metrics, which can display7model-specific performance metrics such as localization error (in mm), prediction confidence (if probabilistic models are used), and QRS window7used. The interface can include export functions, which allow users to export prediction results and visualizations for inclusion in clinical reports or for integration with electroanatomical mapping systems.

[0038] In some embodiments, the system includes a Software implementation functional facility750. The solution can be implemented as a software module, deployable on clinical workstations or cloud-based systems. It can be integrated into existing ECG acquisition platforms or used as a standalone decision-support tool. The entire pipeline is encapsulated in a software platform, allowing clinicians to interact with the model through a graphical user interface. The platform supports real-time analysis, model visualization, and integration into clinical decision-making workflows. The software is designed to assist physicians — particularly cardiac electrophysiologists — by providing real-time or near-realtime feedback during arrhythmia diagnosis and treatment. Once the 12-lead ECG is input into the system 1, the physician can view the predicted arrhythmia origin. The predicted site of arrhythmia origin can be displayed on a 3D anatomical model of the right ventricle. If patient-Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 specific imaging is available, the prediction can shown directly on the personalized heart geometry'. The software platform can allow physicians to interpret clinical relevance. The software can highlight surrounding anatomical landmarks and may optionally overlay structural abnormalities (e.g.. scar zones) from CT / MRI, helping the physician correlate predictions with known pathophysiology7. The software platform can allow physicians to export results. The physician can export coordinates, prediction reports, and visualizations for inclusion in patient records, research documentation, or integration with electroanatomical mapping systems (e.g.. CARTO. EnSite).

[0039] By combining high-resolution ECG analysis with anatomically grounded machine learning, the present disclosure provides a physical (computational) solution to the problem of arrhythmia localization — without requiring invasive mapping catheters or costly 3D mapping systems.

[0040] FIG. IB depicts an exemplary method 100 is depicted. From the above description relating to the system 1 of FIG. 1A, it should be appreciated that the system 1 can be configured to carry7out the method shown in FIG. IB. In some embodiments, the method 100 can include acquiring data (such as patient data) in a step 102, extracting features from the data in a step 104, creating an anatomical model in a step 106. mapping features to spatial coordinates of the anatomical model in a step 108, predicting a region of interest (such as an origin of an arrythmia) in a step 110, and presenting the region of interest to a user in a step 112.

[0041] For instance, in the step 102, patient data can be acquired. The patient data can be ECG data, such as 12-lead ECG data. In some embodiments, the patient data can include cardiac imaging data, such as CT or MRI imaging data, of the patient.

[0042] In some embodiments, in the step 104, features can be extracted from the patient data received in step 102. For example, a 12-lead electrocardiogram (ECG) can be processed to compute QRS integrals (JQRS) from eight independent leads (I, II, V1-V6). These values can optionally be trimmed to identify time-restricted segments of interest (e.g., 30-160 ms), improving predictive precision. In some embodiments, a good segment of interest is one that captures the dominant portion of ventricular depolarization while minimizing the inclusion of noise, baseline drift, or irrelevant post-depolarization components. It can be identified based on either (1) a fixed time window relative to the pacing stimulus or QRS onset (e.g., 30-160 ms), or (2) data-driven optimization, where various time intervals are tested to determine which yields the low est prediction error during model training and validation. The segment may alsoAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 be trimmed dynamically to exclude periods of low signal amplitude or poor signal-to-noise ratio using thresholds derived from the ECG waveform or prior clinical knowledge. In some embodiments, the recorded 12-lead ECG signals are processed to extract quantitative features such as QRS integrals or time-domain waveform characteristics. These features sen e as spatial indicators of electrical activation patterns and form the input for subsequent computational analysis. In some embodiments, the QRS integrals, which can optionally be trimmed, sen e as spatial indicators of electrical activation and form the input for subsequent computational analysis.

[0043] In some embodiments, in the step 106, an anatomical model can be provided to function as a geometric framework onto which patient electrical features (e.g., ECG, QRS data, or trimmed QRS data) are mapped. In some embodiments, a pre-constructed 3D RV mesh model derived from CT imaging (or from other imaging modality) can be used as the anatomical reference frame. In some embodiments, the surface of the RV model is discretized into multiple indiviudal elements, with each centroid representing a candidate pacing or arrhythmia origin site. The RV surface of the model can be discretized into any number of indiviudal elements depending on the size of the CT (or other modality) imaging data set used to generate the 3D RV mesh model. In some embodiments, the pre-constructed 3D RV mesh model can be patient generic. For instance, the 3D RV mesh model can be constructed from imaging data generated from a plurality of patients.

[0044] In some embodiments, a patient specific RV mesh model can be prepared by fitting 3D imaging data gathered of the patient 2 (e.g., of the patient's heart) to the generic, preconstructed 3D RV mesh model (e.g.. the discretized RV surface of the model). For instance, in some embodiments, the medical imaging data (e.g., cardiac imaging data), acquired in the step 102, can be mapped, or fitted, to the pre-constructed, generic, RV mesh model (e.g., using a rigid registration method). It should be appreciated that conversely, the generic, preconstructed model can be mapped, or fitted, to the patient's 3D model (e.g., the data gathered of the patient), for instance by using a rigid registration method.

[0045] In some embodiments, creating the anatomical model in step 106 can include image segmentation, surface mesh generation, and registration and projection. The image segmentation can be of the cardiac chambers, particularly the RV, using thresholding or deep learning-based tool. During surface mesh generation, the segmented volume is converted into a triangulated surface representation using algorithms such as marching cubes or mesh simplification techniques. During registration and projection, the generic model can optionallyAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 be registered onto patient-specific geometry (e g., the imaging data generated of the patient 2) to align spatial coordinate systems, enabling feature transfer or model fusion.

[0046] In some embodiments, in step 108, ECG features (e.g., the ECG data, such as QRS integrals, gathered of the patient 2) are mapped to spatial coordinates (x, y, z) on the RV mesh (e.g., the generic pre-constructed mesh model or the patient’s 3D anatomical model mapped to the generic pre-constructed mesh model).

[0047] While not shown in Fig. IB, it should be appreciated that in some embodiments, the method 100 includes training a model, as discussed above with respect to the machine learning engine functional facility 30, for mapping ECG features to spatial coordinates on the RV mesh model. The trained model can then be used for predicting the origin of a arrythmia in the patient.

[0048] In some embodiments, in step 110, a region of interest can be predicted. In some embodiments, the extracted features are processed to estimate the spatial location of the arrhythmic source. The predicted spatial location can be mapped to the corresponding triangle centroid on the RV mesh (e.g., the generic pre-constructed mesh model or the patient’s 3D anatomical model mapped to the generic pre-constructed mesh model). Given an input ECG (e.g., the ECG data, such as QRS integrals, gathered of the patient 2), a likely origin site on the RV mesh, which can be visualized in a custom software interface (e.g., the generic preconstructed mesh model or the patient’s 3D anatomical model mapped to the generic preconstructed mesh model) can be predicted. For instance, the site or region of interest on the cardiac geometry, indicating the estimated exit site or propagation pathway of the arrhythmia, can be predicted. After a three-dimensional spatial coordinate (x, y. z) corresponding to the likely site of origin is predicted, the predicted coordinate can be mapped to the closest centroid on a generic right ventricular (RV) surface mesh (for instance having 277 triangular elements). If patient-specific RV geometry' is available (e.g., reconstructed from CT or MRI), the predicted site can be automatically projected from the generic model onto the corresponding location on the personalized RV surface. This ensures spatial alignment with the patient’s anatomical structure, enhancing clinical relevance.

[0049] In some embodiments, in the step 112, the predicted region of interest can be visualized in 3D on the anatomical model via a software interface, reported as a coordinate output, and optionally exported for integration with navigation systems or electroanatomical mapping tools. The interface can include a 3D visualization panel, which displays the RV mesh with highlighted regions showing the predicted arrhythmia site. The mesh can be rotated,Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 zoomed, and color-coded to reflect prediction confidence or error distance. The interface can include a coordinate output display, which shows the (x, y, z) coordinate of the predicted site numerically. The interface can include overlay features, which can overlay ground truth sites, error vectors, or scar regions from cardiac imaging, if available. The interface can include prediction metrics, which can display model-specific performance metrics such as localization error (in mm), prediction confidence (if probabilistic models are used), and QRS window used. The interface can include export functions, which allow' users to export prediction results and visualizations for inclusion in clinical reports or for integration with electroanatomical mapping systems.

[0050] The following examples are put forth so as to provide those of ordinary skill in the art with a complete disclosure and description of how' to make and use the assay, screening, and therapeutic methods of the disclosure, and are not intended to limit the scope of w hat the inventors regard as their disclosure.

[0051] Non-Limiting Examples:MethodsStudy Population

[0052] This study included two cohorts. 1) Development Cohort: Sequential, consenting patients undergoing VT catheter ablation were enrolled, and 8 patients underwent standard 12-lead ECG recordings during the procedure (Table 1). 2) Validation Cohort: An independent cohort consisting of 27 RV pacing sites pooled from 3 patients who underwent VT catheter ablation was retrospectively included (Table 1). All patients provided written informed consent prior to enrollment, and the study protocol was approved by the institutional Research Ethics Board.Table 1. Baseline characteristics of patients„.. . .xDevelopment Cohort Validation Cohort Clinical Characteristics (N_8) (N_3)Male (%) 5(62.5) 3Age (years) 47.9±14.9 60.0±7.0 NICM 6 (75.0)RVOT 2ICM 2 (25.0) 1Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 EF (%) 53.6 54ICD present 4 (50.0) 1 (33.3)ECG Acquisition and Processing

[0053] For the 11 patients, 12-lead ECG signals were acquired using the Cardiolab system (GE Healthcare Inc., Barrington, IL). Signals were filtered (0.05-100 Hz), sampled at 1000 Hz with 16-bit resolution, and recorded for 7 seconds. A secured auxiliary' computer received a real-time clone of 8 independent ECG leads (I, II, V1-V6). ensuring no interference with the clinical workflow. For each pacing site, a representative paced beat with Stim-QRS < 40 ms (indicating successful local capture without conduction delay and beat fusion) was selected for analysis. For each selected beat, two ty pes of features were computed: (1) QRS integral (JQRS): Time-integral signal of the QRS complex, measured in microvolt-seconds (pVs); (2) Trimmed fQRS: fQ S computed over shorter time windows, ranging from 30 ms to 160 ms in 10-ms increments, to evaluate how QRS duration affects localization performance.Prototype of the Right Ventricular Endocardium

[0054] A three-dimensional (3D) endocardial surface model of the right ventricle (RV) was reconstructed from a contrast-enhanced CT scan of a structurally normal human heart. The resulting triangulated mesh consists of 277 surface elements, each defined by three vertices in Cartesian coordinates. Fig. 2 shows the corresponding 2D bullseye plot representation of the anatomical model. As shown in Fig. 2, the generic right ventricular endocardial surface is represented in 3D space by 277 triangular elements and projected onto a polar display. The model is divided into 18 anatomical segments: 1. Anterior RVOT, 2. RVOT left septal, 3. RVOT right septal, 4. Posterior RVOT, 5. Lateral RVOT, 6. Apical septal, 7. Apical inferior, 8. Apical anterior, 9. Middle anterior septal, 10. Middle inferior septal, 11. Middle inferior, 12. Middle lateral, 13. Middle anterior, 14. Basal anterior septal, 15. Basal inferior septal 16. Basal inferior, 17. Basal lateral, and 18. Basal anterior.Regression Model with Support Vector Regression (SVR)

[0055] Feature and Target Construction: A quantifiable mapping between ECG features and the spatial locations of pacing sites on the RV surface was hypothesized.

[0056] Input Features: JQRS values from eight ECG leads (I, II, V1-V6), computed over varying durations (30-160 ms).Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026

[0057] Regression Targets: Each pacing site was assigned to the centroid of one of the 277 mesh triangles on the RV model, and the (x, y, z) coordinates of that centroid served as the regression targets.

[0058] Support Vector Regression: Separate SVR models were trained for each coordinate (x, y, z), using two kernel functions: radial basis function (RBF) and linear.

[0059] RBF kernel:

[0060] For each SVR model, the input consisted of an 8-dimensional feature vector derived from JQRS values across eight ECG leads. The target output was one of the spatial coordinates (x, y. or z) of the pacing site, assigned from the centroid of the RV mesh triangle.

[0061] The SVR model maps the ECG feature space to the spatial coordinate using a nonlinear kernel. The similarity between two feature vectors is computed using the RBF kernel:K (m, m1) = exp(— y II m — m1||2) (1),where m and m ’ are ECG feature vectors, and v is a tunable hyperparameter. The predicted coordinate is obtained by a weighted combination of these similarities, based on support vectors learned during training.

[0062] Linear kernel:

[0063] Similarly, linear SVR models were trained to predict each of the three coordinates. The kernel function was defined as the inner product:K(m, m1) = Ef=1m, m} (2)where d is the dimensionality of the input feature vector (i.e., number of ECG features).

[0064] Prior to training, ECG features and coordinate values were standardized using StandardScaler to enhance numerical stability. Hyperparameters including the kernel coefficient (C), epsilon (s) tolerance parameter, and gamma (y) were optimized using a random-search method to minimize localization error (details in Table SI below).Table SI: SVR Hyperparameters ValuesValueHyperparameter - X Y ZEpsilon (E) 0.3 0.2 0.2Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 Gamma (y) 0.2 0.6Kernel Coefficient (C) 1.7 2

[0065] Used in both RBF and Linear kernels, as determined by random-search method, controlling for individual kernels.Model Training, Testing

[0066] The Development Cohort was split into training (80%) and test (20%) subsets.

[0067] Three separate SVR models (one for x. y, z) were trained for each kernel type. The predicted coordinates (x', y', z') were mapped to the closest centroid on the RV mesh using a K-nearest neighbors (KNN) algorithm, ensuring anatomical consistency. Localization error was defined as the Euclidean distance between predicted and ground-truth coordinates, in millimeters.

[0068] QRS Interval Optimization: To identify the optimal QRS integration window, leave-one-out cross-validation was applied across different time intervals (30-160 ms, in 10-ms increments). For each interval, 1000 bootstrapped samples (n ~ 17 held-out samples) were generated, and the mean localization error was computed. The interval with the lowest average error was selected for model training.

[0069] Prediction Accuracy Assessment in the Development Cohort: After identifying the optimal jQRS integration interval, the localization performance of both SVR kernels (linear and RBF) was evaluated using the training and test sets from the Development Cohort. Localization accuracy was quantified as the Euclidean distance (in millimeters) between the predicted coordinates and the ground-truth pacing site.

[0070] Prediction Accuracy Assessment in the Validation Cohort: After identifying the optimal QRS integration interval, both RBF and linear SVR models were retrained on the full Development Cohort. These models were then used in a "black-box" fashion to localize 34 pacing sites in the independent Validation Cohort, without further retraining.Evaluation Metrics

[0071] To quantitatively assess model performance, the following error metrics were computed: Coefficient of determination (2), Root Mean Squared Error (RMSE), and Symmetric Mean Absolute Percentage Error (SMAPE).Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 2” i(7) - Pi)2R2= 1 - ^r=1(Pi - p)2RMSE =A1 V [Ti - Pi]

[0072] Where Ti and Pi are the true and predicted coordinate values, and is the mean of true values, and n is the number of samples.Statistical Analysis

[0073] Paired t-tests were used to compare mean localization errors across QRS intervals from the bootstrapped samples. After identifying the optimal JQRS window, further paired t-tests were conducted to compare RBF and linear SVR performance in both the Development (test subset, n = 45) and Validation Cohorts (n = 34). All tests were conducted at a significance level of a = 0.05.ResultsPatients

[0074] The Development Cohort included eight patients (mean age: 47.9 ± 14.9 years; 62.5% male), while the Validation Cohort included three additional patients (mean age: 60.0 ± 7.0 years; all male) (see Table 1 above). A total of 227 RV pacing sites were collected from the Development Cohort and 27 sites from the Validation Cohort, each associated with corresponding 12-lead ECG data.Identifying an Optimal fQRS Interval

[0075] Referring to Fig. 7, all QRS integration intervals were evaluated using 1000 bootstrapped trials with leave-one-out cross-validation to assess model fit and localization accuracy. Specifically, Fig. 7 is a heatmap of significance in localization accuracy between QRS interval windows. Fig. 7 shows an assessment of optimal JQRS interval between 1000 bootstrapped trial models for the test set for Linear and Gaussian models. Lesser numbers in the table show greater significance, and greater numbers in the table show lesser significance to no significant difference. Shading indicates a magnitude of significance, with lesser numbersAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 in darker shading having the greatest significance, and greater numbers in darker shading have the least significance. Note that the last two rows of the table of Fig, 7 display a comparison between 30 ms vs. 60 ms and 90 ms vs. 120 ms instead of neighboring intervals.

[0076] Among all tested intervals, the 60 ms fQRS window achieved the best overall performance, yielding the lowest localization error (12.0 mm), lowest RMSE and SMAPE, and highest R2 when using the SVR with an RBF kernel. As shown in Figure 3, the RBF SVR consistently outperformed the linear SVR across all tested time intervals. Specifically, Fig. 3 shows localization error in (mm) of arrhythmia origin predictions in the RV-endocardial mesh based on 1000-tnal bootstrapped RBF SVR and linear SVR models for all fQRS interval windows (30 ms, 40 ms, ..., 160 ms) with SD bars.Localization Accuracy Based on the Optimal fQRS Interval

[0077] Using the optimal 60 ms fQRS window, localization performance was evaluated on the Development Cohort. As shown in Table 2(a) below, the RBF SVR model predicting the y-axis coordinate achieved the best performance, with RMSE = 4.590, SMAPE = 0.082%, and R2 = 0.926. In contrast, the x-axis model yielded the weakest performance, with RMSE values of 6.930 (RBF) and 11.252 (linear), and R2 values of 0.778 (RBF) and 0.433 (linear). Similar trends were observed in the independent test set (see Table 2(b) below). The z-axis prediction yielded the highest R2 values (0.858 for RBF, 0.796 for linear), while the x-axis predictions again showed the lowest R2 (0.511 for RBF, 0.215 for linear). Overall, the mean localization error across the test set was 14.4 mm for the RBF kernel and 17.0 mm for the linear kernel. On the training set, the RBF SVR also outperformed the linear SVR (9.5 mm vs. 16.6 mm). These results are summarized in Figure 4. Specifically, Fig. 4 shows localization error in [mm] measured as Euclidean distance of pacing-site origin prediction on the RV-endocardial surface for design set (left panel, n=182) and test set (right panel, n=45) for 60 ms fQRS Interval for RBF and linear methods.Table 2(a): Model Validation Based on the Design SetX axis Y axis Z axis Model RMSE SMAPE R2RMSE SMAPE R2RMSE SMAPE R2Linear 11.252 1.90% 0.433 9.185 0.181% 0.555 10.902 4.28% 0.762 RBF 6.930 1.04% 0.778 4.590 0.082% 0.926 7.546 3.16% 0.893Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026

[0078] Table (2a) shows model performance on the development (design) set, comparing linear and radial basis function (RBF) support vector regression (SVR) models across the x-, y-, and z-coordinate axes. Metrics include root mean squared error (RMSE), symmetric mean absolute percentage error (SMAPE). and coefficient of determination (R2).Table 2(b): Model Validation Based on the Test SetX axis Y axis Z axis Model RMSE SMAPE R2RMSE SMAPE R2RMSE SMAPE R2Linear 12.334 1.99% 0.215 10.067 0.199% 0.442 10.156 4.21% 0.796 RBF 9.863 1.56% 0.511 9.950 0.196% 0.406 8.569 3.22% 0.858

[0079] Table 2(b) shows model performance on the independent test set using the same evaluation metrics and axis-wise comparison as in Table 2(a).Localization Performance in the Validation Cohort

[0080] To further validate the generalizability of our approach, model performance was evaluated on an independent Validation Cohort (n = 34). SVR models were retrained on the entire Development Cohort (n = 227) using the optimal 60 ms fQRS inter al and applied in a black-box fashion to the validation data. While the RBF kernel model again achieved lower localization error than the linear model, the difference was not statistically significant (p = 0.31 ; Table 3). To assess the impact of annotation variability, pacing site coordinates from two independent observers were compared using the trained SVR model. There was a significant difference in the mean localization accuracy between the two observers for the Linear SVR model (p=0.04) and not a significant difference for the RBF SVR model (p=0.22), as shown in Table 4 below.Table 3: Localization Accuracy for Additional Validation Cohort Validation Set (n=34)„ i No. Pacing . . . ,.Case Kernel Mean SD Median _ SitesRBF 9.9 4.8 8.4 1 5Linear 4.9 3.3 5.0 2 RBF 11 11.6 6.9 9.4Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 Linear 11.5 7.4 9.4 RBF 18.3 8.5 18.5 3 18Linear 15.5 10.2 14.8[00811 Table 3 shows localization accuracy for the validation cohort (n = 34), estimated by patient case and kernel type. Performance metrics include the number of pacing sites, mean localization error, standard deviation (SD), and median error (in millimeters) for both radial basis function (RBF) and linear support vector regression (SVR) models.Table 4: Localization Accuracy for Validation Cohort Using Data from Two Observers Observer 1 Observer 2RBF Linear RBF Linear Mean 18.0 17.1 16.4 15.2 SD 8.2 8.1 9.4 9.3 Median 16.5 15.9 14.5 13.8

[0082] Table 4 shows localization performance (in millimeters) for each SVR model as assessed by two independent observers when using the 60 ms JQRS interval. Results are reported for both radial basis function (RBF) and linear kernels. Metrics include mean error, standard deviation (SD). and median error for each observer and model type.Custom Software Demonstration

[0083] To illustrate the application of the custom software for RV pacing site localization, an example case from Procedure #1 in the Validation Cohort is presented. In this demonstration, an unknown pacing-site ECG was recorded (Figure 5A) and processed using eight independent ECG leads as input features. These features were entered into the trained SVR model in a black-box fashion to predict the pacing site location. The predicted site (shown as a black square 202 in Fig. 5b) corresponded to the centroid of one of the 277 triangular elements on the generic RV surface mesh, highlighted by a red star 205 in Figure 5C. That is, the predicted location (black square 202) is shown over the RV mesh in Figure 5B and corresponds to the center of one of the 277 predefined triangular elements (shown as star 205 in Fig. 5C). The linear SVR model localized the site with an accuracy of 4.2 mm, measured asAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 the Euclidean distance between the predicted location (red star 205) and the true pacing-site location (blue star 206 shown in Fig. 5C).

[0084] Techniques operating according to the principles described herein may be implemented in any suitable manner. Included in the discussion above are one or more flow charts showing the steps and acts of various processes that localize right ventricular arrhythmia origins. The processing and decision blocks of the flow charts above represent steps and acts that may be included in algorithms that carry7out these various processes. Algorithms derived from these processes may be implemented as software integrated with and directing the operation of one or more single- or multi-purpose processors, may be implemented as functionally-equivalent circuits such as a Digital Signal Processing (DSP) circuit, Field Programmable Gate Array (FPGA), or an Application-Specific Integrated Circuit (ASIC), or may be implemented in any other suitable manner. It should be appreciated that the flow charts included herein do not depict the syntax or operation of any particular circuit or of any particular programming language or type of programming language. Rather, the flow charts illustrate the functional information one of ordinary skill in the art may use to fabricate circuits or to implement computer software algorithms to perform the processing of a particular apparatus carrying out the types of techniques described herein. It should also be appreciated that, unless otherw ise indicated herein, the particular sequence of steps and / or acts described in each flow chart is merely illustrative of the algorithms that may be implemented and can be varied in implementations and embodiments of the principles described herein.

[0085] Accordingly, in some embodiments, the techniques described herein may be embodied in computer-executable instructions implemented as software, including as application software, system software, firmware, middleware, embedded code, or any other suitable type of software. Such computer-executable instructions may be written using any of a number of suitable programming languages and / or programming or scripting tools, and also may be compiled as executable machine language code or intermediate code that is executed on a framework or virtual machine.

[0086] When techniques described herein are embodied as computer-executable instructions, these computer-executable instructions may be implemented in any suitable manner, including as a number of functional facilities, each providing one or more operations to complete execution of algorithms operating according to these techniques. A "functional facility,” however instantiated, is a structural component of a computer system that, when integrated with and executed by one or more computers, causes the one or more computers toAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 perform a specific operational role. A functional facility may be a portion of or an entire software element. For example, a functional facility may be implemented as a function of a process, or as a discrete process, or as any other suitable unit of processing. If techniques described herein are implemented as multiple functional facilities, each functional facility may be implemented in its own way: all need not be implemented the same way. Additionally, these functional facilities may be executed in parallel and / or serially, as appropriate, and may pass information between one another using a shared memory' on the computer(s) on which they are executing, using a message passing protocol, or in any other suitable way.

[0087] Generally, functional facilities include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. Typically, the functionality of the functional facilities may be combined or distributed as desired in the systems in which they operate. In some implementations, one or more functional facilities carrying out techniques herein may together form a complete software package. These functional facilities may, in alternative embodiments, be adapted to interact with other, unrelated functional facilities and / or processes, to implement a software program application.

[0088] Some exemplary functional facilities have been described herein for carrying out one or more tasks. It should be appreciated, though, that the functional facilities and division of tasks described is merely illustrative of the type of functional facilities that may implement the exemplar}' techniques described herein, and that embodiments are not limited to being implemented in any specific number, division, or type of functional facilities. In some implementations, all functionality may be implemented in a single functional facility. It should also be appreciated that, in some implementations, some of the functional facilities described herein may be implemented together with or separately from others (i.e., as a single unit or separate units), or some of these functional facilities may not be implemented.

[0089] Computer-executable instructions implementing the techniques described herein (when implemented as one or more functional facilities or in any other manner) may, in some embodiments, be encoded on one or more computer-readable media to provide functionality to the media. Computer-readable media include magnetic media such as a hard disk drive, optical media such as a Compact Disk (CD) or a Digital Versatile Disk (DVD), a persistent or non-persistent solid-state memory (e.g., Flash memory', Magnetic RAM, etc.), or any other suitable storage media. Such a computer-readable medium may be implemented in any suitable manner, including as computer-readable storage media 2006 of FIG. 6 describedAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 below (i.e., as a portion of a computing device 2000) or as a stand-alone, separate storage medium. As used herein, “computer-readable media” (also called “computer-readable storage media”) refers to tangible storage media. Tangible storage media are non-transitory and have at least one physical, structural component. In a “computer-readable medium,” as used herein, at least one physical, structural component has at least one physical property that may be altered in some way during a process of creating the medium with embedded information, a process of recording information thereon, or any other process of encoding the medium with information. For example, a magnetization state of a portion of a physical structure of a computer-readable medium may be altered dunng a recording process.

[0090] In some, but not all, implementations in which the techniques may be embodied as computer-executable instructions, these instructions may be executed on one or more suitable computing device(s) operating in any suitable computer system, including the exemplary computer system of FIG. 1. or one or more computing devices (or one or more processors of one or more computing devices) may be programmed to execute the computerexecutable instructions. A computing device or processor may be programmed to execute instructions when the instructions are stored in a manner accessible to the computing device / processor, such as in a local memory (e.g., an on-chip cache or instruction register, a computer-readable storage medium accessible via a bus, a computer-readable storage medium accessible via one or more networks and accessible by the device / processor, etc.). Functional facilities that comprise these computer-executable instructions may be integrated with and direct the operation of a single multi-purpose programmable digital computer apparatus, a coordinated system of two or more multi-purpose computer apparatuses sharing processing power and jointly carrying out the techniques described herein, a single computer apparatus or coordinated system of computer apparatuses (co-located or geographically distributed) dedicated to executing the techniques described herein, one or more Field-Programmable Gate Arrays (FPGAs) for carrying out the techniques described herein, or any other suitable system.

[0091] FIG. 6 illustrates one exemplary implementation of a computing device in the form of a computing device 2000 that may be used in a system implementing the techniques described herein, although others are possible. It should be appreciated that FIG. 6 is intended neither to be a depiction of necessary components for a computing device to operate in accordance with the principles described herein, nor a comprehensive depiction.

[0092] Computing device 2000 may comprise at least one processor 2002, a network adapter 2004, and computer-readable storage media 2006. Computing device 2000 may be, forAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 example, a desktop or laptop personal computer, a personal digital assistant (PDA), a smart mobile phone, a server, a wireless access point or other networking element, or any other suitable computing device. Network adapter 2004 may be any suitable hardware and / or software to enable the computing device 2000 to communicate wired and / or wirelessly with any other suitable computing device over any suitable computing network. The computing network may include wireless access points, switches, routers, gateways, and / or other networking equipment as well as any suitable wired and / or wireless communication medium or media for exchanging data between two or more computers, including the Internet. Computer-readable media 2006 may be adapted to store data to be processed and / or instructions to be executed by one or more processors 2002. Processor 2002 enables processing of data and execution of instructions. The data and instructions may be stored on the computer-readable storage media 2006.

[0093] The data and instructions stored on computer-readable storage media 2006 may comprise computer-executable instructions implementing techniques which operate according to the principles described herein. In the example of FIG. 6, computer-readable storage media 2006 stores computer-executable instructions implementing various facilities and storing various information as described above. Computer-readable storage media 2006 may store the various processes / facilities discussed above. For instance, computer-readable storage media 2006 can store the above-described input feature extraction functional facility 10, anatomical model integration functional facility 20, machine learning engine functional facility730, prediction output functional facility 40, and software implementation functional facility 50.

[0094] While not illustrated in FIG. 6. a computing device may additionally have one or more components and peripherals, including input and output devices. These devices can be used, among other things, to present a user interface. Examples of output devices that can be used to provide a user interface include printers or display screens for visual presentation of output and speakers or other sound generating devices for audible presentation of output. Examples of input devices that can be used for a user interface include keyboards, and pointing devices, such as mice, touch pads, and digitizing tablets. As another example, a computing device may receive input information through speech recognition or in other audible format.

[0095] Embodiments have been described where the techniques are implemented in circuitry and / or computer-executable instructions. It should be appreciated that some embodiments may be in the form of a method, of which at least one example has been provided. The acts performed as part of the method may be ordered in any suitable way. Accordingly,Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 embodiments may be constructed in which acts are performed in an order different than illustrated, which may include performing some acts simultaneously, even though shown as sequential acts in illustrative embodiments.

[0096] Non-limiting embodiments of the present disclosure are set out in the below clauses, which can be combined arbitrarily and, at least, as set out below:

[0097] Clause 1. A system, comprising: at least one processor; and at least one storage medium having encoded thereon executable instructions that, when executed by the at least one processor, cause the at least one processor to carry out a method comprising: receiving surface electrocardiogram (ECG) data of a patient; translating the surface ECG data into 3D spatial coordinates on a right ventricle (RV) anatomical model; and predicting an origin of a ventricular arrhythmia (VA), based on the translated ECG data, on the RV anatomical model.

[0098] Clause 2. The system of clause 1, wherein the RV anatomical model is a 3D model comprising a mesh surface.

[0099] Clause 3. The system of clause 1 or clause 2, wherein the mesh surface is discretized into a plurality of elements, and wherein each centroid of each element represents a candidate pacing site or a candidate arrhythmia origin site.

[0100] Clause 4. The system of any preceding clause, wherein the 3D spatial coordinates are mapped to a closest centroid of an element of the RV anatomical model using a nearest-neighbor or K-nearest-neighbors algorithm.

[0101] Clause 5. The system of any preceding clause, wherein the mesh surface is discretized into at least 250 elements.

[0102] Clause 6. The system of any preceding clause, wherein the RV anatomical model is a patient-generic, pre-constructed RV anatomical model.

[0103] Clause 7. The system of any preceding clause, wherein: the method further comprises receiving imaging data of a heart of the patient; and the RV anatomical model comprises the imaging data mapped to a patient-generic, pre-constructed RV anatomical model.

[0104] Clause 8. The system of any preceding clause, wherein the imaging data is registered to the patient-generic, pre-constructed RV anatomical model using rigid registration, landmark-based registration, surface-based registration, deformable registration, or a combination thereof.

[0105] Clause 9. The system of any preceding clause, wherein predicting the origin of the VA on the RV anatomical model comprises: identifying an optimal QRS interval forAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 accurately capturing RV depolarization in the surface ECG data; and using RV pacing data to localize the origin of the VA on a RV surface of the RV anatomical model.

[0106] Clause 10. The system of any preceding clause, wherein: the surface ECG data comprises 12-lead ECG data; the method further comprises processing the 12-lead ECG data to compute QRS integrals from eight independent leads; the translating the surface ECG data into 3D spatial coordinates on the RV anatomical model comprises translating the QRS integrals into 3D spatial coordinates on the RV anatomical model; and the predicting the origin of the VA, based on the translated ECG data, on the RV anatomical model comprises predicting the origin of the VA, based on the translated QRS integrals, on the RV anatomical model.

[0107] Clause 11. The system of any preceding clause, wherein: the method further comprises trimming the QRS integrals to identify time-restricted segments of interest of the QRS integrals; translating the QRS integrals into 3D spatial coordinates on the RV anatomical model comprises translating the time-restricted segments of interest of the QRS integrals into 3D spatial coordinates on the RV anatomical model; and predicting the origin of the VA, based on the translated QRS integrals, on the RV anatomical model comprises predicting the origin of the VA, based on the time-restricted segments of interest of the QRS integrals, on the RV anatomical model.

[0108] Clause 12. The system of any preceding clause, wherein the method comprises identifying the time-restricted segments of interest of the QRS integrals based on a fixed time window relative to a pacing stimulus or QRS onset.

[0109] Clause 13. The system of any preceding clause, wherein the time-restricted segments of interest of the QRS integrals are identified by evaluating a plurality of candidate time windows and selecting a time window associated with a lowest localization error.

[0110] Clause 14. The system of any preceding clause, wherein the plurality of candidate time windows span 30 ms to 160 ms.

[0111] Clause 15. The system of any preceding clause, wherein the selected time window is about 60 ms from QRS onset.

[0112] Clause 16. The system of any preceding clause, wherein predicting the origin of the VA comprises estimating an exit site or a propagation pathway of the VA.

[0113] Clause 17. The system of any preceding clause, wherein: the RV anatomical model is a 3D model comprising a mesh surface; the mesh surface is discretized into a plurality of elements; each centroid of each element represents a candidate pacing site or a candidate arrhythmia origin site; and predicting the origin of the VA comprises: predicting 3D spatialAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 coordinates of the origin; and mapping the predicted 3D spatial coordinates to a closest centroid on the mesh surface of the RV anatomical model.

[0114] Clause 18. The system of any preceding clause, wherein: the RV anatomical model is a patient-generic, pre-constructed RV anatomical model; and the method further comprises projecting the predicted 3D spatial coordinates mapped to the RV anatomical model to patient-specific RV geometry.

[0115] Clause 19. The system of any preceding clause, wherein: the method further comprises using supervised learning to train a model with at least one of: labelled pacing data; simulated arrythmia events; or a dataset of known pacing sites, wherein each input vector is associated with a known ground truth coordinate on a RV surface; the trained model translates the surface ECG data into the 3D spatial coordinates on the RV anatomical model; and the trained model predicts the origin of the VA, based on the translated ECG data, on the RV anatomical model.

[0116] Clause 20. The system of any preceding clause, wherein: the method further comprises processing the surface ECG data to generate an 8-dimensional feature vector comprising QRS integral values from leads I, II, and V1-V6; the translating the surface ECG data into 3D spatial coordinates on the RV anatomical model comprises translating the 8-dimensional feature vector into 3D spatial coordinates on the RV anatomical model; and the predicting the origin of the VA, based on the translated ECG data, on the RV anatomical model comprises predicting the origin of the VA, based on the 8-dimensional feature vector, on the RV anatomical model.

[0117] Clause 21. The system of any preceding clause, wherein: the method further comprises standardizing the QRS integral values; the 8-dimensional feature vector comprises the standardized QRS integral values; translating the 8-dimensional feature vector into 3D spatial coordinates on the RV anatomical model is performed by a trained model; and predicting the origin of the VA, based on the 8-dimensional feature vector, on the RV anatomical model is performed by the trained model.

[0118] Clause 22. The system of any preceding clause, wherein translating the surface ECG data into 3D spatial coordinates comprises applying separate regression models to predict x-, y-, and z-coordinates of a candidate arrhythmia origin.

[0119] Clause 23. The system of any preceding clause, wherein the method further comprises determining at least one of a confidence value or an error estimate of the predicted origin of the VA on the RV anatomical model.Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026

[0120] Clause 24. The system of any preceding clause, wherein the method further comprises ranking candidate regions for the origin on the RV anatomical model.

[0121] Clause 25. The system of any preceding clause, wherein the method further comprises: displaying the predicted origin on a 3D visualization of the RV anatomical model; and exporting the predicted origin for integration with an electroanatomical mapping system.

[0122] Clause 26. The system of any preceding clause, wherein the predicted origin is displayed together with one or more anatomical landmarks, scar regions, or structural abnormalities derived from CT or MRI data.

[0123] Clause 27. A method, comprising: receiving surface electrocardiogram (ECG) data of a patient; translating the surface ECG data into 3D spatial coordinates on a right ventricle (RV) anatomical model; and predicting an origin of a ventricular arrhythmia (VA), based on the translated ECG data, on the RV anatomical model.

[0124] Clause 28. The method of clause 27, wherein the RV anatomical model is a 3D model comprising a mesh surface.

[0125] Clause 29. The method of clause 27 or clause 28, wherein the mesh surface is discretized into a plurality of elements, and wherein each centroid of each element represents a candidate pacing site or a candidate arrhythmia origin site.

[0126] Clause 30. The method of any preceding clause, wherein the 3D spatial coordinates are mapped to a closest centroid of an element of the RV anatomical model using a nearest-neighbor or K-nearest-neighbors algorithm.

[0127] Clause 31. The method of any preceding clause, wherein the mesh surface is discretized into at least 250 elements.

[0128] Clause 32. The method of any preceding clause, wherein the RV anatomical model is a patient-generic, pre-constructed RV anatomical model.

[0129] Clause 33. The method of any preceding clause, further comprising receiving imaging data of a heart of the patient, wherein the RV anatomical model comprises the imaging data mapped to a patient-generic, pre-constructed RV anatomical model.

[0130] Clause 34. The method of any preceding clause, wherein the imaging data is registered to the patient-generic, pre-constructed RV anatomical model using rigid registration, landmark-based registration, surface-based registration, deformable registration, or a combination thereof.

[0131] Clause 35. The method of any preceding clause, wherein predicting the origin of the VA on the RV anatomical model comprises: identifying an optimal QRS interval forAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 accurately capturing RV depolarization in the surface ECG data; and using RV pacing data to localize the origin of the VA on a RV surface of the RV anatomical model.

[0132] Clause 36. The method of any preceding clause, wherein: the surface ECG data comprises 12-lead ECG data; the method further comprises processing the 12-lead ECG data to compute QRS integrals from eight independent leads; the translating the surface ECG data into 3D spatial coordinates on the RV anatomical model comprises translating the QRS integrals into 3D spatial coordinates on the RV anatomical model; and the predicting the origin of the VA, based on the translated ECG data, on the RV anatomical model comprises predicting the origin of the VA, based on the translated QRS integrals, on the RV anatomical model.

[0133] Clause 37. The method of any preceding clause further comprising trimming the QRS integrals to identify time-restricted segments of interest of the QRS integrals, wherein: translating the QRS integrals into 3D spatial coordinates on the RV anatomical model comprises translating the time-restricted segments of interest of the QRS integrals into 3D spatial coordinates on the RV anatomical model: and predicting the origin of the VA, based on the translated QRS integrals, on the RV anatomical model comprises predicting the origin of the VA, based on the time-restricted segments of interest of the QRS integrals, on the RV anatomical model.

[0134] Clause 38. The method of any preceding clause further comprising identifying the time-restricted segments of interest of the QRS integrals based on a fixed time window relative to a pacing stimulus or QRS onset.

[0135] Clause 39. The method of any preceding clause, wherein the time-restricted segments of interest of the QRS integrals are identified by evaluating a plurality of candidate time windows and selecting a time window associated with a lowest localization error.

[0136] Clause 40. The method of any preceding clause, wherein the plurality of candidate time windows span 30 ms to 160 ms.

[0137] Clause 41. The method of any preceding clause, wherein the selected time window is about 60 ms from QRS onset.

[0138] Clause 42. The method of any preceding clause, wherein predicting the origin of the VA comprises estimating an exit site or a propagation pathway of the VA.

[0139] Clause 43. The method of any preceding clause, wherein: the RV anatomical model is a 3D model comprising a mesh surface; the mesh surface is discretized into a plurality of elements; each centroid of each element represents a candidate pacing site or a candidate arrhythmia origin site; and predicting the origin of the VA comprises: predicting 3D spatialAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 coordinates of the origin; and mapping the predicted 3D spatial coordinates to a closest centroid on the mesh surface of the RV anatomical model.

[0140] Clause 44. The method of any preceding clause, wherein: the RV anatomical model is a patient-generic, pre-constructed RV anatomical model; and the method further comprises projecting the predicted 3D spatial coordinates mapped to the RV anatomical model to patient-specific RV geometry.

[0141] Clause 45. The method of any preceding clause, wherein: the method further comprises: using supervised learning to train a model with at least one of: labelled pacing data; simulated arrythmia events; or a dataset of known pacing sites, wherein each input vector is associated with a known ground truth coordinate on a RV surface; the trained model translates the surface ECG data into the 3D spatial coordinates on the RV anatomical model; and the trained model predicts the origin of the VA, based on the translated ECG data, on the RV anatomical model.

[0142] Clause 46. The method of any preceding clause further comprising processing the surface ECG data to generate an 8-dimensional feature vector comprising QRS integral values from leads I, II, and V1-V6, wherein: the translating the surface ECG data into 3D spatial coordinates on the RV anatomical model comprises translating the 8-dimensional feature vector into 3D spatial coordinates on the RV anatomical model; and the predicting the origin of the VA, based on the translated ECG data, on the RV anatomical model comprises predicting the origin of the VA, based on the 8-dimensional feature vector, on the RV anatomical model.

[0143] Clause 47. The method of any preceding clause further comprising standardizing the QRS integral values, wherein: the 8-dimensional feature vector comprises the standardized QRS integral values; translating the 8-dimensional feature vector into 3D spatial coordinates on the RV anatomical model is performed by a trained model; and predicting the origin of the VA, based on the 8-dimensional feature vector, on the RV anatomical model is performed by the trained model.

[0144] Clause 48. The method of any preceding clause, wherein translating the surface ECG data into 3D spatial coordinates comprises applying separate regression models to predict x-, y-, and z-coordinates of a candidate arrhythmia origin.

[0145] Clause 49. The method of any preceding clause further comprising determining at least one of a confidence value or an error estimate of the predicted origin of the VA on the RV anatomical model.Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026

[0146] Clause 50. The method of any preceding clause further comprising wherein the method further comprises ranking candidate regions for the origin on the RV anatomical model.

[0147] Clause 51. The method of any preceding clause further comprising: displaying the predicted origin on a 3D visualization of the RV anatomical model; and exporting the predicted origin for integration with an electroanatomical mapping system.

[0148] Clause 52. The method of any preceding clause, wherein the predicted origin is displayed together with one or more anatomical landmarks, scar regions, or structural abnormalities derived from CT or MRI data.

[0149] Clause 53. At least one computer-readable storage medium encoded with computer-executable instructions that, when executed by at least one processor, cause the at least one processor to carry' out a method comprising: receiving surface electrocardiogram (ECG) data of a patient; translating the surface ECG data into 3D spatial coordinates on a right ventricle (RV) anatomical model; and predicting an origin of a ventricular arrhythmia (VA), based on the translated ECG data, on the RV anatomical model.

[0150] Various aspects of the embodiments described above may be used alone, in combination, or in a variety of arrangements not specifically discussed in the embodiments described in the foregoing and is therefore not limited in its application to the details and arrangement of components set forth in the foregoing description or illustrated in the drawings. For example, aspects described in one embodiment may be combined in any manner with aspects described in other embodiments.

[0151] All patents and publications mentioned in this specification are herein incorporated by reference to the same extent as if each independent patent and publication was specifically and individually indicated to be incorporated by reference.

[0152] It should be appreciated that the methods discussed above need not be completed in the order of steps discussed above. Multiple steps of the methods can be performed substantially simultaneously or in a different order than presented above. Similarly, multiple steps of the methods can be omitted, or additional steps could be added.

[0153] As utilized herein, the terms “comprise” and “comprising” are intended to be construed as being inclusive, not exclusive. As utilized herein, the terms “exemplary”, “example”, and “illustrative”, are intended to mean “serving as an example, instance, or illustration” and should not be construed as indicating, or not indicating, a preferred or advantageous configuration relative to other configurations. As utilized herein, the terms “about”, “generally”, and “approximately” are intended to cover variations that may existingAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 in the upper and lower limits of the ranges of subjective or objective values, such as variations in properties, parameters, sizes, and dimensions. In one non-limiting example, the terms “about"’, “generally’", and “approximately” mean at, or plus 10 percent or less, or minus 10 percent or less. In one nonlimiting example, the terms “about”, “generally”, and “approximately” mean sufficiently close to be deemed by one of skill in the art in the relevant field to be included. As utilized herein, the term “substantially” refers to the complete or nearly complete extend or degree of an action, characteristic, property7, state, structure, item, or result, as would be appreciated by one of skill in the art. For example, an object that is “substantially” circular would mean that the object is either completely a circle to mathematically determinable limits, or nearly a circle as would be recognized or understood by one of skill in the art. The exact allowable degree of deviation from absolute completeness may in some instances depend on the specific context. However, in general, the nearness of completion will be so as to have the same overall result as if absolute and total completion were achieved or obtained. The use of “substantially” is equally applicable when utilized in a negative connotation to refer to the complete or near complete lack of an action, characteristic, property, state, structure, item, or result, as would be appreciated by one of skill in the art. The use of the terminology X “or” Y herein should be interpreted as meaning either “X” or “Y” individually, or both “X and Y” together.

[0154] Numerous modifications and alternative embodiments of the present disclosure will be apparent to those skilled in the art in view of the foregoing description. Accordingly, this description is to be construed as illustrative only and is for the purpose of teaching those skilled in the art the best mode for carrying out the present disclosure. Details of the structure may vary substantially without departing from the spirit of the present disclosure, and exclusive use of all modifications that come within the scope of any appended claims is reserved. Within this specification, embodiments have been described in a way which enables a clear and concise specification to be written, but it is intended and will be appreciated that embodiments may7be variously combined or separated without parting from the scope of the present disclosure. It is intended that the present disclosure be limited only to the extent required by any appended claims and the applicable rules of law.

Claims

Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 CLAIMSWhat is claimed is:

1. A system, comprising:at least one processor; andat least one storage medium having encoded thereon executable instructions that, when executed by the at least one processor, cause the at least one processor to carry out a method comprising:receiving surface electrocardiogram (ECG) data of a patient;translating the surface ECG data into 3D spatial coordinates on a right ventricle (RV) anatomical model; andpredicting an origin of a ventricular arrhythmia (VA), based on the translated ECG data, on the RV anatomical model.

2. The system of claim 1, wherein the RV anatomical model is a 3D model comprising a mesh surface.

3. The system of claim 2, wherein the mesh surface is discretized into a plurality of elements, and wherein each centroid of each element represents a candidate pacing site or a candidate arrhythmia origin site.

4. The system of claim 3, wherein the 3D spatial coordinates are mapped to a closest centroid of an element of the RV anatomical model using a nearest-neighbor or K-nearest-neighbors algorithm.

5. The system of claim 3, wherein the mesh surface is discretized into at least 250 elements.

6. The system of claim 1, wherein the RV anatomical model is a patient-generic, preconstructed RV anatomical model.

7. The system of claim 1, wherein:the method further comprises receiving imaging data of a heart of the patient; andAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 the RV anatomical model comprises the imaging data mapped to a patient-generic, preconstructed RV anatomical model.

8. The system of claim 7, wherein the imaging data is registered to the patient-generic, preconstructed RV anatomical model using rigid registration, landmark-based registration, surface-based registration, deformable registration, or a combination thereof.

9. The system of any one of claims 1-8, wherein predicting the origin of the VA on the RV anatomical model comprises:identifying an optimal QRS interval for accurately capturing RV depolarization in the surface ECG data; andusing RV pacing data to localize the origin of the VA on a RV surface of the RV anatomical model.

10. The system of any one of claims 1-8, wherein:the surface ECG data comprises 12-lead ECG data;the method further comprises processing the 12-lead ECG data to compute QRS integrals from eight independent leads;the translating the surface ECG data into 3D spatial coordinates on the RV anatomical model comprises translating the QRS integrals into 3D spatial coordinates on the RV anatomical model; andthe predicting the origin of the VA, based on the translated ECG data, on the RV anatomical model comprises predicting the origin of the VA, based on the translated QRS integrals, on the RV anatomical model.

11. The system of claim 10, wherein:the method further comprises trimming the QRS integrals to identify time-restricted segments of interest of the QRS integrals;translating the QRS integrals into 3D spatial coordinates on the RV anatomical model comprises translating the time-restricted segments of interest of the QRS integrals into 3D spatial coordinates on the RV anatomical model; andAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 predicting the origin of the VA, based on the translated QRS integrals, on the RV anatomical model comprises predicting the origin of the VA, based on the time-restricted segments of interest of the QRS integrals, on the RV anatomical model.

12. The system of claim 11, wherein the method comprises identifying the time-restricted segments of interest of the QRS integrals based on a fixed time window relative to a pacing stimulus or QRS onset.

13. The system of claim 11, wherein the time-restricted segments of interest of the QRS integrals are identified by evaluating a plurality of candidate time windows and selecting a time window associated with a lowest localization error.

14. The system of claim 13. wherein the plurality of candidate time windows span 30 ms to 160 ms.

15. The system of claim 13, wherein the selected time window is about 60 ms from QRS onset.

16. The system of any one of claims 1-8, wherein predicting the origin of the VA comprises estimating an exit site or a propagation pathway of the VA.

17. The system of any one of claims 1-8, wherein:the RV anatomical model is a 3D model comprising a mesh surface;the mesh surface is discretized into a plurality of elements;each centroid of each element represents a candidate pacing site or a candidate arrhythmia origin site; andpredicting the origin of the VA comprises:predicting 3D spatial coordinates of the origin; andmapping the predicted 3D spatial coordinates to a closest centroid on the mesh surface of the RV anatomical model.

18. The system of claim 17, wherein:the RV anatomical model is a patient-generic, pre-constructed RV anatomical model; andAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 the method further comprises projecting the predicted 3D spatial coordinates mapped to the RV anatomical model to patient-specific RV geometry.

19. The system of any one of claims 1-8, wherein:the method further comprises using supervised learning to train a model with at least one of:labelled pacing data;simulated arrythmia events; ora dataset of known pacing sites, wherein each input vector is associated with a known ground truth coordinate on a RV surface;the trained model translates the surface ECG data into the 3D spatial coordinates on the RV anatomical model; andthe trained model predicts the origin of the VA. based on the translated ECG data, on the RV anatomical model.

20. The system of any one of claims 1-8, wherein:the method further comprises processing the surface ECG data to generate an 8-dimensional feature vector comprising QRS integral values from leads I, II, and V1-V6; the translating the surface ECG data into 3D spatial coordinates on the RV anatomical model comprises translating the 8-dimensional feature vector into 3D spatial coordinates on the RV anatomical model; andthe predicting the origin of the VA, based on the translated ECG data, on the RV anatomical model comprises predicting the origin of the VA, based on the 8-dimensional feature vector, on the RV anatomical model.

21. The system of claim 20, wherein:the method further comprises standardizing the QRS integral values;the 8-dimensional feature vector comprises the standardized QRS integral values; translating the 8-dimensional feature vector into 3D spatial coordinates on the RV anatomical model is performed by a trained model; andpredicting the origin of the VA, based on the 8-dimensional feature vector, on the RV anatomical model is performed by the trained model.Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 22. The system of any one of claims 1-8, wherein translating the surface ECG data into 3D spatial coordinates comprises applying separate regression models to predict x-, y-, and z-coordinates of a candidate arrhythmia origin.

23. The system of any one of claims 1-8, wherein the method further comprises determining at least one of a confidence value or an error estimate of the predicted origin of the VA on the RV anatomical model.

24. The system of any one of claims 1-8, wherein the method further comprises ranking candidate regions for the origin on the RV anatomical model.

25. The system of any one of claims 1-8, wherein the method further comprises:displaying the predicted origin on a 3D visualization of the RV anatomical model; and exporting the predicted origin for integration with an electroanatomical mapping system.

26. The system of claim 25, wherein the predicted origin is displayed together with one or more anatomical landmarks, scar regions, or structural abnormalities derived from CT or MRI data.

27. A method, comprising:receiving surface electrocardiogram (ECG) data of a patient;translating the surface ECG data into 3D spatial coordinates on a right ventricle (RV) anatomical model; andpredicting an origin of a ventricular arrhythmia (VA), based on the translated ECG data, on the RV anatomical model.

28. The method of claim 27, wherein the RV anatomical model is a 3D model comprising a mesh surface.

29. The method of claim 28, wherein the mesh surface is discretized into a lural, ly of elements, and wherein each centroid of each element represents a candidate pacing site or a candidate arrhythmia origin site.Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 30. The method of claim 29, wherein the 3D spatial coordinates are mapped to a closest centroid of an element of the RV anatomical model using a nearest-neighbor or K-nearest-neighbors algorithm.

31. The method of claim 29, wherein the mesh surface is discretized into at least 250 elements.

32. The method of claim 27, wherein the RV anatomical model is a patient-generic, preconstructed RV anatomical model.

33. The method of claim 27, further comprising receiving imaging data of a heart of the patient, wherein the RV anatomical model comprises the imaging data mapped to a patient-generic, pre-constructed RV anatomical model.

34. The method of claim 33, wherein the imaging data is registered to the patient-generic, preconstructed RV anatomical model using rigid registration, landmark-based registration, surface-based registration, deformable registration, or a combination thereof.

35. The method of any one of claims 27-34, wherein predicting the origin of the VA on the RV anatomical model comprises:identifying an optimal QRS interval for accurately capturing RV depolarization in the surface ECG data; andusing RV pacing data to localize the origin of the VA on a RV surface of the RV anatomical model.

36. The method of any one of claims 27-34, wherein:the surface ECG data comprises 12-lead ECG data;the method further comprises processing the 12-lead ECG data to compute QRS integrals from eight independent leads;the translating the surface ECG data into 3D spatial coordinates on the RV anatomical model comprises translating the QRS integrals into 3D spatial coordinates on the RV anatomical model; andAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 the predicting the origin of the VA, based on the translated ECG data, on the RV anatomical model comprises predicting the origin of the VA, based on the translated QRS integrals, on the RV anatomical model.

37. The method of claim 36 further comprising trimming the QRS integrals to identify time-restricted segments of interest of the QRS integrals, wherein:translating the QRS integrals into 3D spatial coordinates on the RV anatomical model comprises translating the time-restricted segments of interest of the QRS integrals into 3D spatial coordinates on the RV anatomical model; andpredicting the origin of the VA, based on the translated QRS integrals, on the RV anatomical model comprises predicting the origin of the VA, based on the time-restricted segments of interest of the QRS integrals, on the RV anatomical model.

38. The method of claim 37 further comprising identifying the time-restricted segments of interest of the QRS integrals based on a fixed time window relative to a pacing stimulus or QRS onset.

39. The method of claim 37, wherein the time-restricted segments of interest of the QRS integrals are identified by evaluating a plurality of candidate time windows and selecting a time window associated with a lowest localization error.

40. The method of claim 39, wherein the plurality of candidate time windows span 30 ms to 160 ms.

41. The method of claim 39, wherein the selected time window is about 60 ms from QRS onset.

42. The method of any one of claims 27-34, wherein predicting the origin of the VA comprises estimating an exit site or a propagation pathway of the VA.

43. The method of any one of claims 27-34, wherein:the RV anatomical model is a 3D model comprising a mesh surface;the mesh surface is discretized into a plurality of elements;Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 each centroid of each element represents a candidate pacing site or a candidate arrhythmia origin site; andpredicting the origin of the VA comprises:predicting 3D spatial coordinates of the origin; andmapping the predicted 3D spatial coordinates to a closest centroid on the mesh surface of the RV anatomical model.

44. The method of claim 43, wherein:the RV anatomical model is a patient-generic, pre-constructed RV anatomical model; andthe method further comprises projecting the predicted 3D spatial coordinates mapped to the RV anatomical model to patient-specific RV geometry.

45. The method of any one of claims 27-34, wherein:the method further comprises:using supervised learning to train a model with at least one of:labelled pacing data;simulated arrythmia events; ora dataset of known pacing sites, wherein each input vector is associated with a known ground truth coordinate on a RV surface;the trained model translates the surface ECG data into the 3D spatial coordinates on the RV anatomical model; andthe trained model predicts the origin of the VA, based on the translated ECG data, on the RV anatomical model.

46. The method of any one of claims 27-34 further comprising processing the surface ECG data to generate an 8-dimensional feature vector comprising QRS integral values from leads I, II, and V1-V6, wherein:the translating the surface ECG data into 3D spatial coordinates on the RV anatomical model comprises translating the 8-dimensional feature vector into 3D spatial coordinates on the RV anatomical model; andAttorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 the predicting the origin of the VA, based on the translated ECG data, on the RV anatomical model comprises predicting the origin of the VA, based on the 8-dimensional feature vector, on the RV anatomical model.

47. The method of claim 46 further comprising standardizing the QRS integral values, wherein:the 8-dimensional feature vector comprises the standardized QRS integral values; translating the 8-dimensional feature vector into 3D spatial coordinates on the RV anatomical model is performed by a trained model; andpredicting the origin of the VA, based on the 8-dimensional feature vector, on the RV anatomical model is performed by the trained model.

48. The method of any one of claims 27-34, wherein translating the surface ECG data into 3D spatial coordinates comprises applying separate regression models to predict x-. y-. and z-coordinates of a candidate arrhythmia origin.

49. The method of any one of claims 27-34 further comprising determining at least one of a confidence value or an error estimate of the predicted origin of the VA on the RV anatomical model.

50. The method of any one of claims 27-34 further comprising wherein the method further comprises ranking candidate regions for the origin on the RV anatomical model.

51. The method of any one of claims 27-34 further comprising:displaying the predicted origin on a 3D visualization of the RV anatomical model; and exporting the predicted origin for integration with an electroanatomical mapping system.

52. The method of claim 51, wherein the predicted origin is displayed together with one or more anatomical landmarks, scar regions, or structural abnormalities derived from CT or MRI data.Attorney Docket No. 110697-016801 / PCT Electronically Filed: April 17, 2026 53. At least one computer-readable storage medium encoded with computer-executable instructions that, when executed by at least one processor, cause the at least one processor to cany out a method comprising:receiving surface electrocardiogram (ECG) data of a patient;translating the surface ECG data into 3D spatial coordinates on a right ventricle (RV) anatomical model; andpredicting an origin of a ventricular anhythmia (VA), based on the translated ECG data, on the RV anatomical model.

Images