Heart digital twins integrating patient-specific clinical profiles for arrhythmia management and in-silico trials

Abstract

Techniques for providing a personalized 3D heart digital twin of an individual are presented. The techniques include: obtaining 3D imaging data of the individual's heart; generating a 3D heart digital twin geometry of the individual's heart; segmenting the 3D heart digital twin geometry into tissue types; accessing individual clinical profile data representing the individual; passing individual input data to a trained machine learning system, from which individual cellular-level electrophysiological data is obtained, where the trained machine learning system is trained with historical personal clinical profile data sets labeled with respective cellular-level electrophysiological data; assigning, in a 3D heart digital twin based on the 3D heart digital twin geometry, an electrophysiological characteristic to at least the normal tissue type based on the individual cellular-level electrophysiological data; simulating, in the 3D heart digital twin, a future state of the individual; and providing an indication of the future state of the individual.

Description

Attorney Docket No. Pl 8627-02 / 0184.0338-PCTHEART DIGITAL TWINS INTEGRATING PATIENT-SPECIFIC CLINICAL PROFILES FOR ARRHYTHMIA MANAGEMENT AND IN-SILICO TRIALSGovernment Support

[0001] This invention was made with government support under grants HL166759 and HL174440 awarded by the National Institutes of Health. The government has certain rights in the invention.Related Application

[0002] This application claims the benefit of US Provisional Patent Application No. 63 / 773,105, entitled, “Heart Digital Twins Integrating Patient-Specific Clinical Profiles for Arrhythmia Management and In-Silico Trials,” filed March 17, 2025.Field

[0003] This disclosure relates generally to cardiac arrhythmia and the study and treatment thereof.Background

[0004] A digital twin is a virtual representation of a system that mimics the structure and function of the system and has a predictive capability, informing decisions that realize value. Medical digital twins, which are computational models of organs (and even patients) created from patient data, have been making significant inroads in medicine to support clinical decision making. Heart digital twins, which are constructed from clinical and other patient data and represent the functioning of patients’ hearts, have begun to be used for a number of clinical applications in recent years. For instance, atrial digital twins have been proposed as the technology thatcould predict non-invasively, pre-procedure, the optimal personalized targets in patients undergoing atrial fibrillation ablation. However, existing heart digital twins are limited to very few properties, and therefore have restricted predicative power.Summary

[0005] According to various embodiments, a method of providing a personalized three dimensional (3D) heart digital twin of an individual is presented. The method includes: obtaining 3D imaging data of the individual’s heart; generating a 3D heart digital twin geometry of the individual’s heart from the 3D imaging data; segmenting the 3D heart digital twin geometry into a plurality of tissue types, where the plurality of tissue types include a normal tissue type; accessing individual clinical profile data representing the individual; passing individual input data to a trained machine learning system, from which individual cellular-level electrophysiological data is obtained, where the individual input data includes the individual clinical profile data, where the trained machine learning system is trained with a training corpus including a plurality of labeled historical personal clinical profile data sets, where a respective labeled historical personal clinical profile data set is labeled with respective cellular-level electrophysiological data; assigning, in a 3D heart digital twin based on the 3D heart digital twin geometry, an electrophysiological characteristic to at least the normal tissue type based on the individual cellular-level electrophysiological data; simulating, in the 3D heart digital twin, a future state of the individual; and providing an indication of the future state of the individual.

[0006] Various optional features of the above method embodiments include the following. The individual clinical profile data may include genetic data of the individual, drug data of the individual, and disease data of the individual, and a respective labeledhistorical personal clinical profile data set may include respective genetic data, respective drug data, and respective disease data. The individual may include a patient. The simulating may include simulating an ablation of the patient’s heart, where the indication of the future state of the individual is indicative of an ablation outcome for the patient. The simulating may include simulating electrical activity in the patient’s heart, where the indication of the future state of the individual is indicative of a likelihood of sudden cardiac death. The simulating may include simulating a change in a drug administration for the patient, where the indication of the future state of the individual is indicative of the patient’s response to the change in the drug administration for the patient. The simulating may include simulating administering genetic therapy to the patient, where the indication of the future state of the individual is indicative of the patient’s response to the genetic therapy. The simulating may include simulating a change in the patient’s lifestyle, where the indication of the future state of the individual is indicative of the patient’s response to the change in the patient’s lifestyle. The indication of the future state of the individual may include a risk prediction for the patient. The method may include repeating, for a plurality of individuals, the obtaining, generating, segmenting, accessing, passing, and assigning, such that a plurality of 3D heart digital twins for a simulated patient population are obtained. The method may include repeating the simulating for respective 3D heart digital twins of at least the plurality of 3D heart digital twins, such that an in-silico trial on at least the simulated patient population is performed. The simulating may include simulating, in parallel and on a graphical processing unit, electrophysiological wave propagation separately within each of a plurality of voxels.

[0007] According to various embodiments, a non-transitory computer-readable medium including instructions that, when executed by an electronic processor,configure the electronic processor to provide a personalized three dimensional (3D) heart digital twin of an individual by performing the actions of any of the above method embodiments, is presented.

[0008] According to various embodiments, a system for providing a personalized three dimensional (3D) heart digital twin of an individual, the system including an electronic processor and a non-transitory computer-readable medium including instructions that, when executed by the electronic processor, configure the electronic processor to perform the actions of any of the above method embodiments, is presented.

[0009] Combinations, (including multiple dependent combinations) of the above-described elements and those within the specification have been contemplated by the inventors and may be made, except where otherwise indicated or where contradictory.Brief Description of the Drawings

[0010] Various features of the examples can be more fully appreciated, as the same become better understood with reference to the following detailed description of the examples when considered in connection with the accompanying figures, in which:

[0011] Fig. 1 is a schematic diagram of a system for, and method of, generating and using heart digital twins, according to various embodiments;

[0012] Fig. 2 depicts an example of calculating an activation-recovery interval, according to some embodiments;

[0013] Fig. 3 depicts example activation-recovery interval restitution curves for various tissue types, according to some embodiments;

[0014] Fig. 4 depicts personalized action potential curves, according to some embodiments;

[0015] Fig. 5 shows a comparison between predicted and ground-truth action potential parameters, according to an implementation;

[0016] Fig. 6 depicts predicted and true activation-recovery interval restitution curves for various tissue types, according to an implementation;

[0017] Fig. 7 depicts action potential curves for example arrhythmogenic right ventricular cardiomyopathy genotypes;

[0018] Fig. 8 is a flow diagram for a method of providing a personalized three dimensional heart digital twin of an individual, according to some embodiments;

[0019] Fig. 9 illustrates the process used to determine ablation targets for a participant in a clinical trial, according to an embodiment; and

[0020] Fig. 10 illustrates a representative case of a trial participant with a history of a prior coronary artery bypass graft surgery, monomorphic ventricular tachycardia, and prior ventricular tachycardia ablation who presented with recurrent implantable cardioverter-defibrillator shocks.Description of the Examples

[0021] Reference will now be made in detail to example implementations, illustrated in the accompanying drawings. Wherever convenient, the same reference numbers will be used throughout the drawings to refer to the same or like parts. In the following description, reference is made to the accompanying drawings that form a part thereof, and in which is shown by way of illustration specific exemplary examples in which the invention may be practiced. These examples are described in sufficient detail to enable those skilled in the art to practice the invention and it is to beunderstood that other examples may be utilized and that changes may be made without departing from the scope of the invention. The following description is, therefore, merely exemplary.

[0022] Ventricular tachycardia (VT) is a life-threatening arrhythmia leading to sudden cardiac death. Management of VT is difficult, with ablation being a mainstay treatment for VT. However, long-term VT recurrence rates are between 31 % and 48%, often necessitating repeat procedures and patient re-hospitalizations.

[0023] Multiple factors related to the patient’s unique clinical profile, including the distribution of structurally remodeled tissue, medication effects, and genetic arrhythmia predispositions (e.g., arrhythmogenic right ventricular cardiomyopathy (ARVC)), influence VT manifestations and the potential for post-ablation VT recurrence. Various embodiments generate heart digital twins that account for such data, and may be used to determine an appropriate, e.g., optimal, VT treatment procedure that ensures long-term freedom from VT. More generally, various embodiments may be used to provide personalized arrhythmia management approaches, not limited to VT management, that not only account for the comprehensive clinical profile of each patient, but also provide or support personalized clinical decision-making, which improves procedure therapeutic efficacy and minimizes redo ablations and repeat hospitalizations.

[0024] Some embodiments may represent, in a heart digital twin, a wide spectrum of cardiac conditions, such as ischemic and non-ischemic cardiomyopathies, including genetic heart diseases. Various embodiments provide an artificial-intelligence-powered heart digital twin that integrates comprehensive patient-specific clinical profile data, which may be used, for example, for improved ventricular arrhythmia prognostication and treatment. Data used to generate a heart digital twinaccording to some embodiments include any, or any combination, of: clinical imaging modalities (Late Gadolinium Enhanced (LGE)-MRI, Cine-MRI, Contrast Enhanced (CE)-CT, or ultrasound), patient medication history, patient disease history, and genetic profiles, ensuring broad applicability. Some embodiments provide a digital twin platform, which can not only generate a prognosis and / or evaluate a personalized treatment, but also be seamlessly integrated into clinical workflows. Some embodiments provide a paradigm shift in the management of heart dysfunction, with the potential to significantly improve procedural efficacy, reduce recurrence rates, and minimize the need for repeated ablations.

[0025] Some embodiments may provide an entire simulated patient population of heart digital twins. The patient simulated patient population may be subjected to various treatments and / or other changes, such that some embodiments may provide a platform for performing an in-silico clinical trial. For example, some embodiments may serve as a drug development platform.

[0026] Some embodiments use a trained machine learning system to obtain individual cellular-level electrophysiological data based on clinical profile data, and assign electrophysiological characteristic(s) to cardiac tissue types in a 3D heart digital twin based on the individual cellular-level electrophysiological data. The assignment of specific electrophysiological characteristic(s) to particular tissue types based on cellular-level electrophysiological data increases predicative power and accuracy in comparison to prior art techniques that utilize a one-size-fits-all approach to modeling the patient’s baseline cellular-level electrophysiological properties.

[0027] These and other features and advantages are shown and described herein reference to the accompanying figures.

[0028] Fig. 1 is a schematic diagram 100 of a system for, and method of, generating and using one or more heart digital twins, according to various embodiments. The generated digital twin(s) have a variety of uses, e.g., for the precise management of arrhythmia in a patient, and / or for providing a simulated patient population, which may be used to perform in-silico clinical trials, e.g., of the effects of various drugs, genetic therapies, etc. The following description is of generating a single heart digital twin, with the understanding that these actions may be repeated to generate a plurality of heart digital twins.

[0029] At 122, a three-dimensional (3D) contrast-enhanced cardiac scan of an individual is obtained. This data forms the base geometry for the heart digital twin. The individual may be a patient, e.g., for which the generated heart digital twin is used to evaluate treatment options, or the 3D contrast-enhanced cardiac scan may be historic data from an individual other than a current patient, e.g., such that the generated heart digital twin may be used for an in-silico clinical trial. The 3D contrast-enhanced cardiac scan may be a late-gadolinium enhanced magnetic resonance imaging (LGE-MRI) scan, for example. By way of non-limiting example, a cardiac LGE-MRI scan may be obtained using 1.5 Tesla scanners, e.g., available from Avanto, Siemens (Erlangen, Germany) or Signa HDxt, General Electric Medical Systems (Waukesha, Wl), following intravenous injection of gadolinium-based contrast agent with 0.2 mmol / kg dose. By way of non-limiting example, the LGE-MRI scan may have a median of 12 axial slices, with an average slice thickness of 9.33 ± 1.92 mm, and an average in-plane resolution of 1.50 ± 0.38 mm. The scan may be resampled into an isotropic resolution of 0.35 x 0.35 x 0.35 mm, e.g., using the open source software 3D Slicer. Note that embodiments are not limited to LGE-MRI scans. According to some embodiments, the actions of 122 may include acquiring and using a Cine-MRI,Contrast Enhanced (CE)-CT, or ultrasound to provide a base geometry for a heart digital twin.

[0030] At 124, the 3D contrast-enhanced cardiac scan undergoes segmentation. The segmentation may identify myocardium versus non-myocardium regions, and may identify various specific myocardium regions. By way of non-limiting example, biventricular segmentation may be performed using a semiautomatic segmentation method, e.g., as described in E. Sung et al., Fat Infiltration in the Infarcted Heart as a Paradigm for Ventricular Arrhythmias, Nature Cardiovascular Research (2022), available at doi.org / 10.1038 / s44161 -022-00133-6. Both endocardial and epicardial surfaces may be defined with contour points to delineate the myocardial region. Otsu thresholding may be applied to binarize the myocardium into high-intensity and low-intensity regions. In general, the segmentation of 124 may facilitate assigning tissue types and other information to various parts of the heart digital twin.

[0031] At 126, tissue types are assigned to the heart digital twin. Examples of tissue types include: normal tissue, scar tissue, fibrotic tissue, and borderzone tissue. According to some embodiments, dense and diffuse scar / fibrotic tissue may be assigned. A given volume of tissue in the heart digital twin may be assigned one of the possible tissue types. By way of non-limiting example, the mean of lower LGE-MRI signal intensity regions may be used as the reference mean for nonfibrotic myocardium, with, again by way of non-limiting example, a threshold of > 2 standard deviations (SD) of the low-intensity region and < 4 SD used to categorize diffuse fibrosis, and a threshold of > 4 SD used for dense scar.

[0032] Also at 126, according to some embodiments, fiber orientations may be assigned. For example, at each point within the heart digital twin, the Laplace-Dirichletmethod may be used to determine transmural and apicobasal directions. Subsequently, bidirectional spherical linear interpolation may be used to establish fiber orientations based on the fiber orientation rules. Fiber orientations may be assigned to each point in the geometry, to specific tissue types, and / or at discrete locations.

[0033] At 130, cellular-level electrophysiological properties are assigned to the heart digital twin. For example, different electrophysiological properties may be assigned to the different tissue types in the heart digital twin based on the segmentation of 124. Various techniques may be used to facilitate the actions of 130. By way of non-limiting example, this disclosure proceeds to describe, in reference to 102, 104, and 106, a machine learning technique that accommodates a large variety of clinical data, as well as describe, in reference to 108, a separate technique that utilizes genetic data. According to various embodiments, the technique described in reference to 102, 104, and 106 may or may not utilize genetic data. According to various embodiments, where the technique described in reference to 102, 104, and 106 does not utilize genetic data, the technique described in reference to 108 may be used together with the technique described in reference to 102, 104, and 106.

[0034] The technique of 102, 104, and 106 is described in detail presently, in reference to Fig. 1, and also in reference to Figs. 2-6. Briefly, a machine learning system (e.g., a multi-output partial least square regressor, a neural network, a random forest, etc.) is trained to input clinical profile data and output corresponding action potential parameters, e.g., action potential durations, which are then used to characterize electrophysiological properties of one or more tissue types (e.g., a normal tissue type). To train the machine learning system, a corpus of training data is generated. The training data includes clinical profile data sets. Each clinical profile data set is derived from anonymized historical medical records of a person. Thecorresponding electrophysiological data for the person that is used as the label is derived from the person’s unipolar electrogram (UEGs) signals. These data provide the cycle length of the pacing / heart rate during a clinical procedure when the UEG signals were acquired. The UEG signals also provide the corresponding activationrecovery interval (ARI). The cycle length and ARI information is then used to calculate the diastolic interval, using the formula DI = CL - ARI. Plotting ARI against diastolic interval provides ARI restitution curves, from which the cellular-level electrophysiological parameters (e.g., action potential parameters) that are used as labels are derived. The process summarized in this paragraph, which concerns 102 and 104 of Fig. 1, is described in detail herein in reference to Figs. 2-6.

[0035] At 102, clinical profile data is obtained. Fortraining the machine learning system of 104, the clinical profile data is from anonymized historical medical records. For generating a new digital twin of an individual, the clinical profile data is for the individual. In general, the clinical profile data may be conceptualized as belonging in one or more categories. Example categories are described presently. A first category of clinical profile data is demographic data. Demographic data may include, by way of non-limiting example, one or more of: individual age, individual sex, individual ethnicity, etc. A second category of clinical profile data is drug data. Drug data may include, by way of non-limiting example, one or more of: indications of whether, and dosage of, medications such as ACE inhibitors and / or amiodarone. A third category of clinical profile data is disease data. Disease data may include, by way of nonlimiting example, indications of whether the individual has: hypertension, atrial fibrillation, diabetes, COPD, coronary artery disease, hypercholesterolemia, and / or chronic kidney disease. A fourth category of clinical profile data is quantitative late gadolinium enhanced cardiac magnetic resonance imaging (LGE-CMR) data. LGE-CMR data may include, by way of non-limiting example, one or more of: fibrosis amount (e.g., percent), border zone amount (e.g., percent), and / or scar tissue amount (e.g., percent). A fifth category of clinical profile data is quantitative electrocardiogram (ECG) data. Quantitative ECG data may include, by way of non-limiting example, one or more of: heart rate, QRS value, QT value, and / or QTc value. A sixth category of clinical profile data is genetic data. Genetic data may include, by way of non-limiting example, indications of whether the individual has a arrhythmogenic right ventricular cardiomyopathy (ARVC) plakophilin-2 (PKP2) phenotype, an ARVC genetic-elusive (GE) genotype, or neither. Other clinical profile data, not limited to that expressly presented herein, may be used according to various embodiments.

[0036] A description of deriving training data from historical medical record clinical profile data follows.

[0037] Fig. 2 depicts an example 200 of calculating an activation-recovery interval from a UEG signal, according to some embodiments. The UEG signal may be from an intra-procedural electroanatomical mapping acquired during a ventricular tachycardia ablation substrate mapping. The activation-recovery interval may be extracted from the UEG, which may include sinus rhythm, right ventricular (RV) pacing, and RV extrastimulus pacing. In general, the activation-recovery interval is used as a surrogate for action potential duration. As shown in Fig. 2, the activation-recovery interval may be calculated from the UEG signal, by way of non-limiting example, as the interval between the steepest negative slope (depolarization time) and the steepest positive slope (repolarization time), which may be represented as ARI = RT - T, by way of non-limiting example. Once a plurality of activation-recovery interval values are obtained, they are processed as shown and described herein in reference to Fig. 3.

[0038] Fig. 3 depicts example activation-recovery interval restitution curves 300 for various tissue types, according to some embodiments. To obtain the activationrecovery interval restitution curves 300, the activation-recovery intervals are processed and plotted as functions of the preceding diastolic interval (DI), where the DI is the difference between the cycle length and the activation-recovery interval, and where the cycle lengths correspond to different pacing rates (60000ms / heart rate for sinus rhythm). These data may be collected for different tissue types; Fig. 3 depicts action-recovery interval restitution curves 300 for border zone and normal tissue types. The activation-recovery interval restitution curve for normal tissue is based on a UEG voltage threshold of > 8.3mV.

[0039] Regressions are then fit to the activation-recovery interval restitution curves 300, e.g., using a geometry-constrained loss function, to calculate the cellular-level action potential parameters, which are then used in ionic models that characterize the electrophysiological properties of the various tissue types. According to some embodiments, the cellular-level electrophysiological parameters, which characterize how cells behave across different cycle lengths, are determined as the constants in the ARI = f(DI) relationship determined by the regression. These cellular-level electrophysiological parameters may be used as the labels on the clinical profile data for training the machine learning system. This process may be repeated for the different tissue types.

[0040] Once the labels are generated, the machine learning system is trained using the labeled historical medical clinical profile data to predict cellular-level electrophysiological parameters from a new individual’s (e.g., a patient’s) clinical profile data. Any of a variety of machine learning systems may be trained, and their hyperparameters tuned, e.g., using a randomized grid search method. Five-fold cross-validation may be used to select a machine learning system that perform best. Thus, once the machine learning system is trained, it may be used to predict regional ionic model electrophysiological properties in the digital twin of an unseen individual.

[0041] Fig. 4 depicts personalized action potential curves 400, according to some embodiments. In particular, the action potential curves 400 represent cellular-level electrophysiological properties of normal tissue and border zone tissue types. The action potential curves 400 were generated by a machine learning trained with a training corpus generated as described herein based on clinical profile data for a novel patient.

[0042] For a study, the inventors trained and tested a machine learning system as described herein. The data for the training corpus was obtained from patients undergoing ventricular tachycardia ablation. Seventeen patients (mean age 68.5 years, 82.4% ICM, 17.6% NICM) underwent LGE-CMR and substrate mapping during ablation for scar-dependent ventricular tachycardia. EAMs we acquired at sinus rhythm, RV pacing at 600ms and during extrastimulus pacing at 20ms above the effective refractory period. UEGs were recorded using a high-density mapping catheter (HD Grid). Using 56,138 UEGs, the machine learning system (a multi-output partial least square regressor) was trained to predict the time constants in the ionic model (tau1 and tau2) that characterize cell behavior across different cycle lengths. Comparisons of predictions with the ground truth values are shown in Fig. 5 for different regions.

[0043] Fig. 5 shows a comparison 500 between predicted and ground-truth action potential parameters tau1 and tau2, according to an implementation. In the comparison 500, dots denote healthy tissue and crosses denote border zone tissue.As is seen from the interpolated line of slope one, the predicted values correlate with the action values for the cellular-level action potential parameters tau1 and tau2.

[0044] The study also tested the machine learning system by comparing its predicted activation-recovery interval restitution curves to activation-recovery interval restitution curves derived directly from the electroanatomical mappings, as shown in Fig. 6.

[0045] Fig. 6 depicts predicted and true activation-recovery interval restitution curves 600 for various tissue types, according to an implementation. In particular, Fig. 6 depicts predicted and actual activation-recovery interval restitution curves for both normal tissue and border zone tissue. The close coherence between the predicted and actual curves demonstrates very good predictive capabilities of the trained machine learning system used in the study.

[0046] Returning to the description of Fig. 1 , once an individual’s clinical profile data is obtained at 102, it is input to a machine learning system 104 trained as described above in reference to Figs. 2-6. The trained machine learning system outputs cellular-level electrophysiological parameters at 106, e.g., action potential parameters, such as tau1 and tau2. At 130, these parameters are then used to assign cellular-level electrophysiological properties to one or more tissue types of the digital twin for the individual.

[0047] As an alternative or complimentary technique to that of 102, 104, and 106, a description of a technique for assigning electrophysiological properties to various tissue types in the heart digital twin based on genetic data is described presently in reference to 108.

[0048] At 108, genetic information for the individual is provided. The genetic information may include an indication of whether the individual has an ARVC PKP2phenotype, an ARVC GE genotype, or neither. This data may be used to generate distinct cellular-level electrophysiological models. For example, the electrophysiological properties of non-fibrotic myocardium may be modeled using distinct cell models for patients with GE and PKP2 phenotypes. The models may include specific ion channel modifications as follows. The human ventricular model from T. O'Hara, etal., Simulation of the Undiseased Human Cardiac Ventricular Action Potential: Model Formulation And Experimental Validation, PLoS ComputBiol 7, e1002061 (2011) may be supplemented with a late sodium current formulation, and this baseline model may be modified to reflect the pathological ionic remodeling caused by PKP2 loss-of-fu notion mutations, for example. Fig. 7 illustrates action potential curves produced using examples of such models.

[0049] Fig. 7 depicts action potential curves 700 for example arrhythmogenic right ventricular cardiomyopathy genotypes. As shown in Fig. 7, the ionic channel adjustments resulted in a PKP2 action potential with a slower upstroke velocity, lower peak, and more positive resting membrane potential than the baseline cell action potential. Consequently, the non-fibrotic myocardium in the PKP2 digital twins may be configured to have slower conduction velocity and a flattened restitution curve, indicating poorer action potential duration adaptation to rapid pacing. For regions of diffuse fibrosis, because the differences in tissue electrophysiological properties between GE and PKP2 are poorly understood, a previously validated version of the TT2 model with channel modifications based on experimental data of non-ischemic cardiomyopathy may be used. This diffuse fibrosis cell model may exhibit a longer action potential duration and slower conduction velocity than both GE and PKP2 models.

[0050] Having assigned cellular-level electrophysiological properties of one or more tissue types to the digital twin at 130, either through the process described herein in reference to 102, 104, and 106, through the process of 108, or by a combination of such processes, a personal heart digital twin is produced at 140. The personal heart digital twin may represent a wide spectrum of cardiac conditions, such as ischemic and non-ischemic cardiomyopathies, including genetic heart diseases. Any combination of the preceding actions may be repeated to generate a plurality of heart digital twins. The heart digital twin(s) may be used for a variety of arrhythmia management processes, e.g., for an individual a patient, and / or for performing an in-silico clinical trials.

[0051] At 150, the personal heart digital twin of 140 is used for personalized arrhythmia management for a patient. According to the actions of 150, the clinical profile data of 102 may be for the patient whose arrhythmia is managed at 150. In general, the heart digital twin may be used to obtain a prognosis or a personalized treatment for the patient. By way of non-limiting example, the personalized heart digital twin may be used to simulate an ablation procedure, to determine what are the optimal location(s) in the heart that, when ablated, would eliminate wavefront reentry. These processes may be integrated seamlessly into various clinical workflows. Further, the personal heart digital twin may account for the patient’s response to any of a variety of actual or potential treatments.

[0052] The actions of 162 are optional, and include generating further heart digital twins in addition to those generated directly from patient medical records. For example, the actions of 162 may include perturbing parameters from an existing personal heart digital twin to generate a synthetic heart digital twin. The parameters may be perturbed randomly, or may be obtained from a different personal heart digitaltwin. According to some embodiments, two or more heart digital twins may be merged, e.g., by sampling parameters from them, averaging parameters from them, or extrapolating parameters from them, to generate a synthetic heart digital twin. Examples of contemplated parameters include: values of any of the clinical profile data of 102, shapes of regions of the segmentation of 124, values of the assigned tissue types of 126, and / or any of the cellular-level electrophysiological properties of 130.

[0053] The actions of 180 (and the actions of optional 162) include using generated personal heart digital twins, including those generated directly from patient data and / or synthesized heart digital twins, as one or more virtual cohorts for an in-silico clinical trial. The in-silico clinical trial may evaluate the effectiveness of any of a variety of procedures, treatments, and / or change, including, by way of non-limiting example, ablation, gene therapy, lifestyle changes, drug therapy, drug changes (e.g., changes to dosage and / or active compound). For example, the generated personal heart digital twins may serve as a drug development platform.

[0054] Further examples of applications for the heart digital twins of 130 and / or 162 are shown and described herein in reference to Fig. 8.

[0055] Fig. 8 is a flow diagram for a method 800 of providing a personalized three dimensional heart digital twin of an individual, according to some embodiments. The method 800 may be practiced using the system and method shown and described herein in reference to Fig. 1, for example.

[0056] At 802, the method 800 includes obtaining 3D imaging data of the individual’s heart. The individual may be a current patient or an individual represented by historical medical record data, for example. The 3D imaging data may be an LGE-MRI, a Cine-MRI, a Contrast Enhanced (CE)-CT, or ultrasound, for example.

[0057] At 804, the method 800 includes generating a 3D heart digital twin geometry of the individual’s heart from the 3D imaging data. The actions of 804 may include sampling or resampling the 3D imaging data, for example. According to some embodiments, the actions of 804 include reformatting the 3D imaging data to a form suitable for subsequent actions.

[0058] At 806, the method 800 includes segmenting the 3D heart digital twin geometry into a plurality of tissue types. The plurality of tissue types may include one or more of: normal tissue, scar tissue, fibrotic tissue, and / or borderzone tissue. According to some embodiments, the plurality of tissue types include one or both of dense and diffuse scar / fibrotic tissue.

[0059] At 808, the method 800 includes accessing individual clinical profile data representing the individual. The individual clinical profile data may include demographic data, genetic data of the individual, drug data of the individual, and / or disease data of the individual. Examples of such data are shown and described in reference to Fig. 1, reference 102.

[0060] At 810, the method 800 includes passing input data for the individual to a trained machine learning system. The input data includes at least the individual clinical profile data. As a consequence of passing the input data, the trained machine learning system outputs individual cellular-level electrophysiological data. The trained machine learning system is trained with a training corpus including a plurality of labeled historical personal clinical profile data sets, e.g., as shown and described herein in reference to Fig. 1, reference 104. In the training corpus, each labeled historical personal clinical profile data set is labeled with respective cellular-level electrophysiological data. According to some embodiments, each labeled historicalpersonal clinical profile data set includes respective genetic data, respective drug data, and respective disease data.

[0061] At 812, the method 800 includes assigning, in a 3D heart digital twin based on the 3D heart digital twin geometry, an electrophysiological characteristic to at least one tissue type based on the individual cellular-level electrophysiological data. A respective electrophysiological characteristic may be assigned to one or more of the tissue types of the segmentation from 806.

[0062] At 814, the method 800 includes simulating, in the 3D heart digital twin, a future state of the individual’s heart. Any of a variety of simulations may be performed per 814. According to some embodiments, the simulating may include simulating an ablation of the patient’s heart, e.g., at one or more identified locations, and the simulation may produce an indication of an ablation outcome for the patient. According to some embodiments, the simulating may include simulating a change in a drug administration for the patient, and the simulation may produce an indication of the patient’s response to the change in the drug administration for the patient. According to some embodiments, the simulating may include simulating administering genetic therapy to the patient, and the simulation may produce an indication of the patient’s response to the genetic therapy. According to some embodiments, the simulating may include simulating a change in the patient’s lifestyle, and the simulation may produce an indication of the patient’s response to the change in the patient’s lifestyle. Note that embodiments are not limited to a single simulation. For example, an embodiment may be capable of performing, and may perform multiple of the above simulations, and / or other simulations.

[0063] According to some embodiments, the simulating may include simulating, in parallel and on a graphical processing unit, electrophysiological wave propagation.The simulated electrophysiological wave propagation may be used to simulate any of a variety of future states of the individual’s heart, non-limiting examples of which follow. According to some embodiments, the simulating may include simulating ablation, and the simulated electrophysiological wave propagation may be used to determine whether the simulated ablation stops all wave reentry. According to some embodiments, the simulating may include simulating a change in a drug (e.g., a change in dosage and / or in drug choice), and the simulated electrophysiological wave propagation may be used to determine how the drug change affects the individual’s heartbeat. According to some embodiments, the simulated electrophysiological wave propagation may be used to determine a probability or likelihood of future sudden cardiac death in the individual. Note that according to some embodiments, simulating electrophysiological wave propagation may be used for a patient, e.g., to develop an individual treatment plan. According to some embodiments, simulating electrophysiological wave propagation may be used in virtual cohorts for an in-silico trial, e.g., to assess the response of the virtual cohort to some intervention, drug, etc.

[0064] The electrophysiological wave propagation may be performed separately within each of a plurality of voxels. For example, some embodiments may utilize the Lattice-Boltzmann method (LBM) to perform the electrophysiological wave propagation simulation. In general, LBM is a pseudo-particle solver that can solve computational fluid dynamic problems or other systems of equations. In LBM, the spatial domain of the system of equations may be discretized into a regular lattice grid of volumetric units, referred to as “voxels.” According to some embodiments that use LBM, the macroscopic function being solved for a system of equations is represented as a very large collection of microscopic particles represented by distribution functions moving from voxel to voxel. Specifically, within each voxel, the function being solvedmay be split into multiple weighted moments distributed in the direction of each face and the center of the voxel. Within the voxel, the governing system of equations may be solved, and the updated moments then streamed (moved) to the neighboring voxel in the direction of the moment. Afterwards, the post-streaming moments may be summed together to calculate the macroscopic solution within each voxel at a discretized point in time. This process may be repeated for each time point until the simulation is completed. Because these processes may be solved within each voxel separately, the entire process can be solved for each voxel separately, allowing for massive parallelization in the solver. Moreover, each computation within an individual voxel is relatively small, so the entire process can run on a graphics processing unit (GPU), resulting in over a 10O-fold speed up in contrast to a comparable finite-element solver ran on a CPU. In tests of an example LBM implementation, equivalent wave propagation simulations took over an hour for a finite element solver, as compared to under one minute with the LBM implementation. Thus, embodiments that utilize LBM may achieve ultra-fast solution times using a single GPU in a desktop computer, rendering them capable of being included in a clinical workflow and deployable across clinical centers.

[0065] At 816, the method 800 includes providing an indication of the future state of the individual. For example, the indication of the future state of the individual may include a risk prediction (e.g., a survival analysis and / or recurrence probability) for the patient.

[0066] According to some embodiments, the method 800 may include repeating, for a plurality of individuals, the actions of one or more of 802, 804, 806, 808, 810, and / or 812. This process may produce a plurality of 3D heart digital twins, representing a simulated patient population (e.g., a virtual cohort). The method 800may also include repeating the simulating actions of 814 and 816, for the simulated patient population. Thus, some embodiments may utilize the simulated patient population to perform an silico trial on the simulated patient population.

[0067] Trial

[0068] The remainder of this application discloses a ten-patient FDA-approved clinical trial on digital twins for ablation of ventricular tachycardia. The trial prospectively tested an embodiment’s ability to guide ischemic VT ablation procedures. The trial successfully demonstrated the embodiment’s safety, feasibility, and excellent clinical outcomes. A brief summary immediately follows, after which details of the trial are presented.

[0069] The trial involved one US institution and ten subjects. Participants underwent contrast-enhanced cardiac MRI to generate a cardiac digital twin reflecting the distribution of structural remodeling. Rapid pacing established VT circuits, whose critical components were then targeted for virtual ablation to render the digital twin non-inducible (details are shown and described in reference to Fig. 9). The digitaltwin-predicted targets were imported into the electroanatomical mapping system to guide clinical ablation. VT was induced via programmed stimulation, and radiofrequency ablation was performed to each digital twin ablation location. Following ablation of all digital twin targets, repeat VT induction was performed. The primary endpoint of VT non-inducibility was adjudicated.

[0070] Baseline characteristics are summarized in the Table 1 below. Eight of ten participants had inducible VT, and all underwent successful digital-twin-guided ablation. Seven participants underwent programmed stimulation following ablation of digital twin targets. Post-ablation programmed stimulation was not performed in one participant due to hemodynamic instability concerns. Six of the remaining sevenpatients were rendered non-inducible following ablation of digital twin targets. The one participant with inducible VT following digital twin-guided ablation in the left ventricle septum was non-inducible for VT following ablation on the RV septum at the digital twin target (details are shown and described in reference to Fig. 10). All ten participants were non-inducible for any VT at the end of the procedure. There were no periprocedural complications.FOLLOW-UP Participant Age Sex LVEF Hypertension Diabetes Prior VT NonFollowATP ICD Death Cause Heart Ablation Induced Inducible up Shock of Transplant before Post Days Death Ablation TWIN-VTAblation1 72 Female 45 No No No Yes Yes 909 No No No - No 2 85 Male 25 Yes Yes Yes Yes Yes 316 No No Yes Cancer No 3 68 Male 30 No Yes Yes Yes Not 251 No No No Yes tested(VF,CPR)4 61 Male 40 Yes No No No - 720 No No No - No 5 77 Male 25 Yes Yes Yes Yes No (RV 480 Yes No No - No ablationneeded)6 68 Male 35 Yes No Yes Yes Yes 349 No No No - No 7 75 Male 35 Yes No Yes Yes Yes 342 No No No - No 8 75 Male 65 Yes No No No - 318 Yes No No - No 9 67 Female 40 No Yes Yes Yes Yes 279 No No No - No 10 61 Male 35 No No Yes Yes Yes 90 No No No - NoTable 1: Base ine Demographics and Procedural Outcomes

[0071] As shown in Table 1 , over a mean follow-up period of 405.4 days (range 90-909 days), eight of ten participants remained VT-free off anti-arrhythmic drug therapy. Of the two participants with VT recurrence, both experienced one VT episode that terminated with ATP within one month of ablation. They remain VT-free with deescalated anti-arrhythmic drug therapy, as summarized in Table 2. No participant received an implantable cardioverter-defibrillator (ICD) shock.Participant Amiodarone-Pre Sotalol-Pre Mexiletine-Pre Amiodarone-Post Sotalol-Post Mexiletine-Post Ablation Ablation Ablation Ablation Ablation Ablation 1 No No No No No No2 Yes No Yes No No No3 Yes No No No No No4 Yes No No No No No5 Yes No Yes Yes (dose No No reduced)6 No No No No No No7 No No No No No No8 Yes No No No Yes No9 Yes No No No No No10 Yes No No No No NoTable 2: Antiarrhythmic Drug Therapy Pre and Post Ablation

[0072] A detailed description of the trial follows immediately below in reference to Figs 9 and 10.

[0073] Trial: Cardiac MRI Protocol

[0074] Cardiac magnetic resonance (CMR) studies for personalized digital twin construction were performed 4-8 weeks before the catheter ablation using a 1.5 Tesla MRI scanner (Aera; Siemens Healthineers, Erlangen, Germany). The CMR protocol included contrast-enhanced MR angiography (CE-MRA) and 2D and 3D late gadolinium enhancement (LGE)-CMR scans. Time-resolved CE-MRA for assessmentof heart anatomy was acquired using the TWIST (Siemens Healthineers, Erlangen, Germany) pulse sequence during intravenous injection of 0.2 mmol / kg of gadobutrol (Bayer Healthcare Pharmaceuticals, Montville, NJ). The typical scan parameters for the TWIST scan were as follows: coronal imaging volume, repetition time of 2.23 ms, echo time of 0.92 ms, flip angle of 25 degrees, in-plane resolution 0.96x1.92 mm, slice thickness 3.0 mm, and reconstructed voxel size of 0.96x0.96x1.5 mm. A 3D LGE-MRI scan to assess the scar and gray zone was initiated 20-25 minutes after the contrast administration. The scan was performed using a 3D fast spoiled gradient recalled echo pulse sequence with wide-band inversion recovery T1 contrast preparation, respiratory navigation, and ECG-gating. The typical scan parameters for a 3D LGE-MRI scan were transverse imaging volume covering the whole heart, TR / TE = 3.5 / 1.38 ms, flip angle = 12 degrees, isotropic spatial resolution 1.5x1.5x1.5 mm, and reconstructed voxel size of 0.75x0.75x0.75 mm. The trigger time and data acquisition duration (the number of segments) were optimized to acquire imaging data during late diastole as defined by inspection of the cine images. The optimal inversion time (Tl) for LGE scans was selected using a Tl-scout scan.

[0075] Trial: Generation of the Heart Digital Twin and Ablation Targeting

[0076] For each participant in the trial, a heart digital twin was generated to characterize the electrical functioning of the participant’s heart based on the participant’s 3D LGE-CMR images. The heart digital twin was assessed for possible VTs through iterative rounds of in-silico VT induction with virtual rapid pacing, analogous to clinical VT induction. VT wave propagation was then analyzed, and in-silico VT ablations performed leading to optimum ablation lesion sets that terminated all VTs the patient’s heart digital twin could sustain. This process is illustrated in Fig. 9,which shows the baseline and additional simulation rounds and the corresponding VT examples and ablation targets, as well as the final optimum ablation strategy plan.

[0077] Fig. 9 illustrates the process used in the trial to determine ablation targets for a participant, according to an embodiment. An example of induced In-silico VT through rapid pacing protocol in the participant’s heart digital twin is shown in the top row. The VT wave front is followed at different time points as shown by straight arrow. Ablation targets were determined that terminated this VT (primary targets). Repeated VT induction in the additional round of simulations, incorporating virtual ablation at the targets identified in the baseline simulation round, revealed an emergent VT shown in the bottom row (view from the endocardium with the scar rendered transparent in the VT wave propagation figures). Additional ablation targets were determined for the termination of this emergent VT (secondary targets). This process was repeated until the participant’s heart digital twin became non-inducible for VT leading to the final (optimal) ablation strategy plan.

[0078] Trial: Heart Digital Twin Geometrical Reconstruction

[0079] The 3D LGE-CMR images were resampled into short axis slices at an isotropic resolution of 0.35 x 0.35 mm, and the endocardial and epicardial ventricular boundaries were semi-automatically segmented using the variational implicit method. The left ventricular myocardium was then reconstructed from the segmentations and was classified, based on signal intensity, as dense scar, borderzone, or healthy tissue using a validated full-width half-maximum approach. Finite element tetrahedral left ventricle mesh was generated using adaptive meshing process in Mimics software (Materialize NV, Leuven, Belgium). Fiber orientations were assigned to the mesh using a rule-based method.

[0080] Trial: Heart Digital Twin Electrophysiological Properties

[0081] Electrophysiological properties were assigned to each of the three myocardial classes in the participant’s heart digital twin. The dense scar was considered electrically non-conductive. Healthy and borderzone were represented by the Ten Tusscher human ventricular ionic model. Electrophysiological remodeling in the borderzone was represented by the modification to the ionic model, resulting in a longer action potential (AP) duration, decreased AP upstroke velocity, decreased AP peak amplitude and decreased conduction velocity. The same cell and tissue electrophysiological properties parameters reported in previous studies were used in this study. See Waight MC, Prakosa A, Li AC, et al. Personalized Heart Digital Twins Detect Substrate Abnormalities in Scar-Dependent Ventricular Tachycardia. Circulation 2025;151 (8):521— 33; Waight MC, Prakosa A, Li AC, et al. Heart Digital Twins Predict Features of Invasive Reentrant Circuits and Ablation Lesions in Scar-Dependent Ventricular Tachycardia. Circ Arrhythm Electrophysiol 2025; 18(8):e013660; and Sung E, Prakosa A, Zhou S, et al. Fat infiltration in the infarcted heart as a paradigm for ventricular arrhythmias. Nat Cardiovasc Res 2022;1 (10):933— 45.

[0082] Trial: VT Induction in the Heart Digital Twin

[0083] Potential VTs in the heart digital twin were assessed by performing in-silico VT inducibility tests using a previously validated rapid pacing protocol. Pacing was delivered as a train of 6 stimuli at 600 ms basic cycle length and followed by up to three successive premature extrastimuli to induce VT. Pacing was performed independently at seven left ventricle sites, each site representing different left ventricle regions based on a condensed 17-segment AHA model, to reveal the possible VTs that the heart digital twin substrate can sustain. The pacing sites were selected to be next to borderzone to maximize the likelihood of inducing VT.

[0084] Trial: Ablation Simulation at Baseline

[0085] VT inducibility test from each pacing site in the heart digital twin was analyzed for reentrant wave propagation. For an identified reentry location, the VT was analyzed for critical sites, including the entrance, isthmus, and exit. Simulated ablation lesions were applied to terminate reentry at the VT location. Each in-silico lesion was represented by an electrically non-conductive region with a radius of 3.5 mm. An extension of the ablation lesion was performed to connect to adjacent regions of non-conducting transmural dense scar, anatomical barriers (such as the mitral anulus), or prior digital twin ablation lesions to avoid the emergence of secondary, iatrogenic VTs following ablation. Lesions applied to the original substrate during this baseline round of simulation were labeled “primary targets.”

[0086] Trial: Repeat Simulations and Additional Ablations

[0087] In-silico VT induction protocol from the seven pacing sites was repeated following the application of ablation lesions. These simulations were performed to assess VT inducibility in the post-ablation substrate, including the residual native scar and ablation lesions, as well as to confirm the elimination of previously observed VTs. Additional ablation lesions were applied to the heart digital twin for any new emergent VTs and were labeled additional or secondary targets. The process was repeated until the participant’s heart digital twin was non-inducible of any VT. When needed, the digital-twin predicted ablation lesion set could also incorporate short linear ablations connecting targeted sites to prevent reentry around ablation lesions, thus decreasing the probability of emergent arrhythmias post-ablation. The final predicted targets comprised the primary and secondary target ablation lesions, from all rounds of simulations, that terminated all in-silico VTs. This is designed to prevent theemergence of new VTs after the index ablation procedure, thus decreasing redo procedures and re-hospitalization.

[0088] Trial: Electrophysiology Study and Catheter Ablation

[0089] Antiarrhythmic medications were discontinued for at least five half-lives before the procedure (aside from amiodarone, which was discontinued two weeks prior). ICD tachyarrhythmia detection was disabled before ablation and restored to the original mode after the procedure. General anesthesia or conscious sedation was used according to clinical requirements. Endocardial electroanatomical mapping of the ventricles was performed during sinus rhythm or ventricular pacing using the CARTO system (Biosense Webster, Inc). Anatomical reconstruction was performed using a high-density multipolar Decanav, Octoray, or Optrell (Biosense Webster) and CARTOSound with an intracardiac echocardiogram catheter.

[0090] The predicted targets needed to be imported into the electroanatomical mapping system to guide the ablation using the virtual predicted ablation targets. To do so, the trial extracted the predicted targets from the digital twin as surface meshes in Virtualization Toolkit (VTK, Kitware Inc.) format for import compatibility with the CARTO electroanatomic mapping system. To register the virtual ablation targets to the participant’s heart during the clinical procedure, corresponding anatomic landmarks from the participant’s digital twin geometric model and electroanatomic mapping anatomic reconstruction were used for the registration process. Landmark registration was performed using the CartoMerge application within the electroanatomical navigation system. The following surfaces were extracted from the participant’s digital twin for use as registration landmarks: left ventricular endocardium, left ventricular apex, right ventricular endocardium, epicardium, and infarct surfaces. Additional surfaces segmented and used as landmarks included the aorta, the leftaortic cusp, and the right aortic cusp. All digital twin surfaces were exported as VTK files and imported into the CARTO system at the beginning of the clinical ablation procedure. The registration process performed with CartoMerge first aligned the coordinates of the left ventricular apex and left and right aortic cusps of the digital twin model to the coordinates of the left ventricular apex and left and right aortic cusps created with the intra-cardiac ultrasound catheter (CartoSound, Biosense Webster Inc.) during the procedure. This landmark-based registration superimposed the digital twin landmarks and ablation targets, to the participant's heart.

[0091] High-density voltage and sinus rhythm / pacing mapping were performed to identify areas of late potentials and isochronal crowding. Ventricular tachycardia was induced via programmed stimulation from the right or left ventricle. For hemodynamically stable VT, activation mapping was performed to identify the VT reentrant circuit. Pace mapping was performed to determine the VT exit sites for hemodynamically unstable VT.

[0092] Ablation was then performed using an irrigated radiofrequency (Thermocool or QDOT, Biosense Webster) catheter. At least 40 watts for 60 seconds was applied to each digital-twin predicted ablation lesion location while monitoring changes in impedance with a target of > 10 ohms decrease or inability to capture at 10 mA at a pulse width of 2 ms. Following the ablation of all the predicted ablation targets, ventricular programmed stimulation was performed with up to triple extrastimuli until ventricular refractoriness or a coupling interval of 200 ms. At this time the primary endpoint was adjudicated. Further ablation was performed at the operator’s discretion.

[0093] Trial: Follow-Up

[0094] The primary endpoint of the study was conversion from acute inducibility of VT to acute non-inducibility of VT using only the lesions indicated by the embodiment. The secondary endpoints were VT requiring ATP or ICD shock therapy in follow-up. The study end date was three months post the last participant ablation procedure. Statistical analysis was performed using StataSE 18.

[0095] Trial: Representative Case

[0096] Fig. 10 illustrates a representative case of a trial participant with a history of a prior coronary artery bypass graft surgery, monomorphic VT, and prior VT ablation who presented with recurrent implantable cardioverter-defibrillator shocks. In particular, Fig. 10 illustrates an ECG 1002 of the trial participant, showing induced VT with a left bundle morphology. Fig. 10 also shows a digital twin 101 Oof the left ventricle with ablation targets (e.g., 1012), around an area of scar (1014). An electroanatomical voltage map of the left ventricle 1020 shows decreased voltage in the basal septum. The digital-twin ablation targets are shown by the arrows. Following ablation (represented as medium-grey spheres, e.g., 1022), the participant remained inducible for the same VT, with the ECG pattern suggestive of a right ventricular exit. Sinus rhythm activation mapping 1030 of the RV showed normal myocardial voltage and no areas of delayed or fractionated tissue, but a 94% pace match to the clinical VT 1032 on the RV septum with isochronal crowding on the LV side of the septum 1034. Combined right and left ventricle ablation (e.g., 1042) rendered the participant noninducible for VT at the end of the procedure, as shown at 1040.

[0097] Certain examples can be performed using a computer program or set of programs. The computer programs can exist in a variety of forms both active and inactive. For example, the computer programs can exist as software program(s) comprised of program instructions in source code, object code, executable code orother formats; firmware program(s), or hardware description language (HDL) files. Any of the above can be embodied on a transitory or non-transitory computer readable medium, which include storage devices and signals, in compressed or uncompressed form. Exemplary computer readable storage devices include conventional computer system RAM (random access memory), ROM (read-only memory), EPROM (erasable, programmable ROM), EEPROM (electrically erasable, programmable ROM), flash memory, and magnetic or optical disks or tapes.

[0098] Aspects of the present disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the disclosure. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented using computer readable program instructions that are executed by an electronic processor.

[0099] These computer readable program instructions may be provided to a processor of a general-purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the electronic processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, a programmable data processing apparatus, and / or other devices to function in a particular manner, such that the computer readable storage medium having instructions stored therein comprises anarticle of manufacture including instructions which implement aspects of the function / act specified in the flowchart and / or block diagram block or blocks.

[0100] In embodiments, the computer readable program instructions may be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, statesetting data, configuration data for integrated circuitry, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++, or the like, and procedural programming languages, such as the C programming language or similar programming languages. The computer readable program instructions may execute entirely on a user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server.

[0101] As used herein, the terms “A or B” and “A and / or B” are intended to encompass A, B, or {A and B}. Further, the terms “A, B, or C” and “A, B, and / or C” are intended to encompass single items, pairs of items, or all items, that is, all of: A, B, C, {A and B}, {A and C}, {B and C}, and {A and B and C}. The term “or” as used herein means “and / or.”

[0102] As used herein, language such as “at least one of X, Y, and Z,” “at least one of X, Y, or Z,” “at least one or more of X, Y, and Z,” “at least one or more of X, Y, or Z,” “at least one or more of X, Y, and / or Z,” or “at least one of X, Y, and / or Z,” is intended to be inclusive of both a single item (e.g., just X, or just Y, or just Z) and multiple items (e.g., {X and Y}, {X and Z}, {Y and Z}, or {X, Y, and Z}). The phrase “at least one of and similar phrases are not intended to convey a requirement that each possible item must be present, although each possible item may be present.

[0103] The techniques presented and claimed herein are referenced and applied to material objects and concrete examples of a practical nature that demonstrably improve the present technical field and, as such, are not abstract, intangible or purely theoretical. Further, if any claims appended to the end of this specification contain one or more elements designated as “means for [perform]ing [a function]...” or “step for [performing [a function]...”, it is intended that such elements are to be interpreted under 35 U.S.C. § 112(f). However, for any claims containing elements designated in any other manner, it is intended that such elements are not to be interpreted under 35 U.S.C. § 112(f).

[0104] While the invention has been described with reference to the exemplary examples thereof, those skilled in the art will be able to make various modifications to the described examples without departing from the true spirit and scope. The terms and descriptions used herein are set forth by way of illustration only and are not meant as limitations. In particular, although the method has been described by examples, the steps of the method can be performed in a different order than illustrated or simultaneously. Those skilled in the art will recognize that these and other variations are possible within the spirit and scope as defined in the following claims and their equivalents.

[0105] Combinations, (including multiple dependent combinations) of the above-described elements and those within the specification have been contemplated by the inventors and may be made, except where otherwise indicated or where contradictory.

Claims

What is claimed is:

1. A method of providing a personalized three dimensional (3D) heart digital twin of an individual, the method comprising:obtaining 3D imaging data of the individual’s heart;generating a 3D heart digital twin geometry of the individual’s heart from the 3D imaging data;segmenting the 3D heart digital twin geometry into a plurality of tissue types, wherein the plurality of tissue types comprise a normal tissue type;accessing individual clinical profile data representing the individual; passing individual input data to a trained machine learning system, from which individual cellular-level electrophysiological data is obtained, wherein the individual input data comprises the individual clinical profile data, wherein the trained machine learning system is trained with a training corpus comprising a plurality of labeled historical personal clinical profile data sets, wherein a respective labeled historical personal clinical profile data set is labeled with respective cellular-level electrophysiological data;assigning, in a 3D heart digital twin based on the 3D heart digital twin geometry, an electrophysiological characteristic to at least the normal tissue type based on the individual cellular-level electrophysiological data;simulating, in the 3D heart digital twin, a future state of the individual; and providing an indication of the future state of the individual.

2. The method of claim 1 , wherein the individual clinical profile data comprises genetic data of the individual, drug data of the individual, and diseasedata of the individual, and wherein a respective labeled historical personal clinical profile data set comprises respective genetic data, respective drug data, and respective disease data.

3. The method of claim 1 , wherein the individual comprises a patient, and wherein the simulating comprises simulating an ablation of the patient’s heart, wherein the indication of the future state of the individual is indicative of an ablation outcome for the patient.

4. The method of claim 1 , wherein the individual comprises a patient, and wherein the simulating comprises simulating electrical activity in the patient’s heart, wherein the indication of the future state of the individual is indicative of a likelihood of sudden cardiac death.

5. The method of claim 1 , wherein the individual comprises a patient, and wherein the simulating comprises simulating a change in a drug administration for the patient, wherein the indication of the future state of the individual is indicative of the patient’s response to the change in the drug administration for the patient.

6. The method of claim 1 , wherein the individual comprises a patient, and wherein the simulating comprises simulating administering genetic therapy to the patient, wherein the indication of the future state of the individual is indicative of the patient’s response to the genetic therapy.

7. The method of claim 1 , wherein the individual comprises a patient, and wherein the simulating comprises simulating a change in the patient’s lifestyle, wherein the indication of the future state of the individual is indicative of the patient’s response to the change in the patient’s lifestyle.

8. The method of claim 1 , wherein the individual comprises a patient, and wherein the indication of the future state of the individual comprises a risk prediction for the patient.

9. The method of claim 1 , further comprising repeating, for a plurality of individuals, the obtaining, generating, segmenting, accessing, passing, and assigning, whereby a plurality of 3D heart digital twins for a simulated patient population are obtained.

10. The method of claim 9, further comprising repeating the simulating for respective 3D heart digital twins of at least the plurality of 3D heart digital twins, whereby an in-silico trial on at least the simulated patient population is performed.

11. The method of claim 1 , wherein the simulating comprises simulating, in parallel and on a graphical processing unit, electrophysiological wave propagation separately within each of a plurality of voxels.

12. A system for providing a personalized three dimensional (3D) heart digital twin of an individual, the system comprising: an electronic processor; and a non-transitory computer-readable medium comprising instructions that, whenexecuted by the electronic processor, configure the electronic processor to perform actions comprising:obtaining 3D imaging data of the individual’s heart;generating a 3D heart digital twin geometry of the individual’s heart from the 3D imaging data;segmenting the 3D heart digital twin geometry into a plurality of tissue types, wherein the plurality of tissue types comprise a normal tissue type;accessing individual clinical profile data representing the individual; passing individual input data to a trained machine learning system, from which individual cellular-level electrophysiological data is obtained, wherein the individual input data comprises the individual clinical profile data, wherein the trained machine learning system is trained with a training corpus comprising a plurality of labeled historical personal clinical profile data sets, wherein a respective labeled historical personal clinical profile data set is labeled with respective cellular-level electrophysiological data;assigning, in a 3D heart digital twin based on the 3D heart digital twin geometry, an electrophysiological characteristic to at least the normal tissue type based on the individual cellular-level electrophysiological data;simulating, in the 3D heart digital twin, a future state of the individual; and providing an indication of the future state of the individual.

13. The system of claim 12, wherein the individual clinical profile data comprises genetic data of the individual, drug data of the individual, and disease data of the individual, and wherein a respective labeled historical personal clinicalprofile data set comprises respective genetic data, respective drug data, and respective disease data.

14. The system of claim 12, wherein the individual comprises a patient, and wherein the simulating comprises simulating electrical activity in the patient’s heart, wherein the indication of the future state of the individual is indicative of a likelihood of sudden cardiac death.

15. The system of claim 12, wherein the individual comprises a patient, and wherein the simulating comprises simulating a change in a drug administration for the patient, wherein the indication of the future state of the individual is indicative of the patient’s response to the change in the drug administration for the patient.

16. The system of claim 12, wherein the individual comprises a patient, and wherein the simulating comprises simulating administering genetic therapy to the patient, wherein the indication of the future state of the individual is indicative of the patient’s response to the genetic therapy.

17. The system of claim 12, wherein the individual comprises a patient, and wherein the simulating comprises simulating a change in the patient’s lifestyle, wherein the indication of the future state of the individual is indicative of the patient’s response to the change in the patient’s lifestyle.

18. The system of claim 12, further comprising repeating, for a plurality of individuals, the obtaining, generating, segmenting, accessing, passing, andassigning, whereby a plurality of 3D heart digital twins for a simulated patient population are obtained.

19. The system of claim 18, further comprising repeating the simulating for respective 3D heart digital twins of at least the plurality of 3D heart digital twins, whereby an in-silico trial on at least the simulated patient population is performed.

20. The system of claim 12, wherein the simulating comprises simulating, in parallel and on a graphical processing unit, electrophysiological wave propagation separately within each of a plurality of voxels.

Images