Systems and methods for cardiac diagnostics

Abstract

A computer-implemented method for generating a cardiac activation map is provided. The method comprises receiving data for at least one cardiac cycle obtained from ECG data of a patient and determining at least one parameter representing cardiac activation based on the received data. The method further comprises obtaining from a database at least one activation map, wherein the obtained at least one activation map is selected from the database based on the determined at least one parameter. The method further comprise generating an initial activation map based on the at least one activation map obtained from the database and generating a final activation map by optimizing the initial activation map.

Description

[0001] SYSTEMS AND METHODS FOR CARDIAC DIAGNOSTICS

[0002] The present application generally relates to systems and methods for cardiac diagnostics, and more particularly to techniques for analyzing and interpreting cardiac data using computer-based algorithms, and to a system for merging multi-modality data for therapy planning and outcome monitoring in cardiac electrophysiology.

[0003] Reference is made to the earlier patent applications DE102024114031.3 of 20 May 2024 and EP25177220.8 of 19 May 2025, the priorities of which are herewith claimed, and the contents of which are herein incorporated by reference.

[0004] Cardiovascular arrhythmia are the leading cause of death in the western world. Even though some mechanisms of disease progression are known, optimal treatment approaches are non-existent for complex arrhythmias such as atrial fibrillation or some non-inducible polymorphic ventricular tachycardias. This is due to the fact that genesis of these diseases is a complex interplay between genetic factors and life-style resulting in highly unpredictable cardiac substrate changes and abnormalities in the excitation patterns.

[0005] Cardiological care may be broadly divided into monitoring and preventive care before and after the procedure on the one hand and acute treatment on the other hand. The former in general is ensured by practitioners during regular check-ups and in acute cases. Treatment is usually delivered either in the form of identifying and acutely ablating arrhythmogenic areas in the heart in the catheter laboratory (cathlab) or / and implantable devices based on the patient's diagnosis and individual characteristics of rhythm conduction and cardiac anatomy.

[0006] Invasive systems monitoring electrical activity of the heart can provide only local information, which is not sufficient for diagnosis of complex non-repeatable patterns of cardiac arrhythmia. Surface-ECG allows qualitative estimation of global conduction patterns. However, this is without 3D coordinates, and lacks sufficient resolution on the local scale.

[0007] Anatomical maps of the cardiac myocardium are sometimes very useful per se, but do not deliver any information on the electrical functionality of the heart. Existing solutions for creating 3 D-el ectroanatom i cal maps or panoramic - i.e., multichamber epi- and endocardial - visualization of the electrical activity of the heart based on non-invasive mapping depend on the use of specific electrode vests or electrode stripes. The number of electrodes is in the range of 160-200 electrodes. The whole thorax of the patient is covered with electrodes. I.e., placement of e.g. defibrillation patches or other sensors / actors is difficult. Location of the electrodes has to be visualized by 3D imaging. A patient individual model of the heart has to be acquired by CT or MRI.

[0008] EP3157423A4 shows a method that involves collecting three-dimensional image data and electrical data from a patient's heart, and then combining these to create a computational model that reflects the heart's time-varying electrical dynamics.

[0009] US20210193291 A1 shows a computer-based system and process are disclosed for reconstructing the internal electrical behavior of a patient's heart based partly or wholly on the patient's electrocardiogram (ECG).

[0010] EP3319521B1 shows a computer-implemented method for creating a map showing how electrical signals travel through the heart.

[0011] US20220369930A1 shows systems and methods for identifying cardiac arrhythmia target segments for non-invasive ablation therapy, particularly stereotactic body radiotherapy (SBRT).

[0012] US20170330075A1 shows systems and methods for personalizing a cardiac electrophysiology (EP) model using deep learning.

[0013] US20210244341 A1 shows systems and methods for localizing arrhythmias, particularly ventricular arrhythmias such as premature ventricular contractions (PVCs) or ventricular tachycardia (VT), using ECG data combined with selected 3D heart models. This application addresses the problem of improving cardiac diagnostic analysis by integrating advanced algorithms and multi-modality imaging data into existing clinical workflows without disrupting the workflow process.

[0014] The application provides solutions for enhancing the accuracy and efficiency of cardiac diagnostic analysis by employing advanced algorithms to calculate the propagation of electrical excitation across the heart using standard 12-lead ECG data. The method further includes constructing personalized 3D models of the excitation spreading of the heart based on the calculated excitation, supplemented by CT, MRI, and / or ultrasound imaging data. Real-time analysis of the heart models is provided, allowing healthcare professionals to access and utilize the analysis seamlessly within their existing clinical workflows. Additionally, the method incorporates deep learning networks for automatic segmentation of cardiac structures from the multi-modality imaging data. Panoramic electroanatom ical maps of the heart calculated by means of analytical methods. The application also enables the integration of invasive electrophysiological data during cardiac procedures, further augmenting the personalized 3D heart models. Overall, this application addresses the need for a comprehensive and integrated approach to cardiac diagnostic analysis, improving patient care and outcomes.

[0015] The present application uses clinical data from surface ECGs, and the results are shown or evaluated in maps. In other words, the present application aims for a patient specific model that closely represents a given surface ECG episode.

[0016] The present application is customized for individual patients, by using a mathematical solution that may be tailored for individual patient conditions.

[0017] The present application represents electroanatomical maps in both, ventricular and atrial chambers of the heart.

[0018] The present application enables panoramic electroanatomical mapping using standard 12-lead ECG as the sole primary input. This approach bypasses the complexity of multimodal imaging, providing a cost-effective, readily available input modality for full 3D cardiac activation analysis. The present application generates dynamic, patient-specific 3D heart models by integrating ECG-derived excitation propagation with multimodal imaging data, thus delivering a more physiologically meaningful model.

[0019] The present application supports real-time cardiac analysis without requiring changes to existing clinical workflows, ensuring seamless adoption and reducing training overhead.

[0020] The present application employs deep learning-based architectures, such as ll-Net based architectures optimized for cardiac anatomy, for automated and robust segmentation of cardiac structures from multimodal imaging, ensuring consistent performance and enabling automation.

[0021] The present application produces comprehensive electroanatomical maps, including activation, propagation, phase, and rotor density maps, derived computationally from ECG data, providing actionable diagnostic insights.

[0022] The present application enhances non-invasive models with selectively incorporated invasive electrophysiological data, achieving a higher fidelity in diagnostics and treatment planning.

[0023] The present application includes a modular cloud platform or web-based access for real-time access to diagnostic outputs and dynamic models, enabling distributed collaboration and remote expert consultation.

[0024] The present application processes multimodal imaging data in combination with ECGbased propagation analysis to improve detection and modeling of fibrotic and scar tissue, thereby enhancing the clinical relevance and precision of electroanatomical representations.

[0025] The present application employs a finite element method to model the relationship between body-surface ECG signals and internal cardiac electrical sources, offering a mathematically rigorous approach to inverse problem solving. The present application utilizes standard 12-lead electrocardiogram (ECG) recordings as the sole input modality for constructing three-dimensional electroanatomical representations. By eliminating the need for specialized electrode vests typically used in electrocardiographic imaging (ECGI), the invention enables streamlined diagnostic procedures that can be performed within existing clinical infrastructure, thereby facilitating broader adoption in routine cardiac care settings.

[0026] The present application is designed for integration into existing clinical workflows without necessitating any procedural modifications. It performs real-time analysis and visualization based on conventional diagnostic inputs, ensuring that clinicians can obtain advanced electrophysiological insights with minimal operational disruption, thus enhancing diagnostic throughput and user adoption.

[0027] The present application constructs personalized three-dimensional heart models by integrating electrical excitation data derived from ECG input with anatomical references. This approach goes beyond static anatomical modeling, delivering functionally tailored representations of the patient’s cardiac physiology and thereby supporting more precise and individualized diagnostic assessments.

[0028] The present application provides a suite of electroanatomical visualizations including but not limited to activation maps, phase maps, and rotor density maps. These representations offer clinicians interactive, interpretable diagnostic views that reveal both the spatial and temporal dynamics of cardiac excitation, thereby supporting refined clinical decision-making in electrophysiological evaluations.

[0029] The present application is architected to function with a variety of imaging and ECG data sources, including both pre-defined anatomical templates and patient-specific input data. This technical flexibility ensures compatibility with standard diagnostic equipment and allows clinicians to apply the system in diverse clinical contexts, regardless of institutional imaging protocols or hardware configurations.

[0030] The present application integrates selectively obtained invasive electrophysiological data into the non-invasive mapping process, to increase diagnostic fidelity. This hybrid data augmentation enhances both local and global model accuracy, allowing the generated electroanatomical representations to more accurately reflect the patient’s electrophysiological state and thereby supporting targeted intervention strategies.

[0031] The present application is implemented as a web-based system that supports real-time interaction and distributed access. This architecture enables remote diagnostic review and collaborative clinical decision-making across different geographic locations, thereby enhancing scalability and promoting consistent care delivery regardless of locationspecific resource availability.

[0032] The present application enhances anatomical accuracy by utilizing multi-modality imaging data, such as CT, MRI, and ultrasound, to construct patient-specific 3D models of the heart. This enables tailored cardiac simulations aligned with each patient's unique structural characteristics.

[0033] The present application supports direct incorporation into established clinical workflows without requiring changes. This facilitates seamless, real-time diagnostic support and decision-making during routine procedures, increasing clinical adoption and usability.

[0034] The present application enables computation of detailed panoramic electroanatomical maps, including activation, propagation, phase, and rotor density visualizations, from ECG-based models. These maps provide a comprehensive view of cardiac electrical dynamics and support advanced diagnostics.

[0035] The present application enriches diagnostic models by incorporating invasive electrophysiological data acquired during or after procedures. Integration of catheter positions, impedance data, and electrograms improves the precision and relevance of real-time electroanatomical maps.

[0036] The present application discloses a web-based platform ensuring that analysis results and visualizations are accessible in real time through standard browsers. This feature supports distributed clinical teams and enhances collaboration in patient care.

[0037] The present application employs a finite element solver to compute internal cardiac excitation sources from body-surface ECG recordings. This inversion process enables model personalization directly from observed ECG data, enhancing diagnostic specificity.

[0038] The present application utilizes standard 12-lead ECG signals as the central input for deriving 3D maps of cardiac excitation propagation. This approach capitalizes on widely available data, making advanced electroanatomical mapping accessible and practical in diverse clinical settings.

[0039] The present application discloses the feature of near-real-time diagnostics based on standard clinical equipment such as a 12-lead ECG enabling seamless integration into routine clinical workflows. This facilitates the adoption of the system in everyday hospital operations without requiring separate or specialized simulation environments.

[0040] The present application discloses the feature of producing interpretable, function-based outputs such as rotor maps or propagation paths enabling immediate therapeutic relevance. This supports clinicians in therapy planning through actionable visualizations, reducing reliance on complex interpretation of raw simulation data.

[0041] The present application discloses the feature of refining patient-specific electrophysiological models using intra-procedural signal data enabling the system to improve diagnostic precision during an intervention. This supports a personalized treatment approach by dynamically adapting the model based on live electrophysiological feedback.

[0042] The present application discloses the feature of web-based access to diagnostic outputs enabling distributed use of the system across multiple clinical departments and locations. This supports collaborative diagnosis and treatment planning, aligning with contemporary digital healthcare infrastructure.

[0043] The present application discloses the feature of solving the inverse problem using a finite element solver to derive cardiac internal activity from observed ECGs enabling a physiologically grounded reconstruction of electrical activity. This represents a clinically aligned modeling approach, as it begins from observable patient data rather than simulated training outputs. The present application utilizes data from multiple imaging modalities, such as CT, MRI, and ultrasound, to construct highly individualized 3D models of the heart. This approach allows the anatomical model to more accurately reflect patient-specific structures and variations, supporting precise diagnostics and planning.

[0044] The present application automatically segments cardiac structures from imaging data using deep learning networks. This ensures a high degree of anatomical precision and significantly reduces manual effort, facilitating more efficient and consistent model generation.

[0045] The present application discloses models which are designed for real-time analysis and are seamlessly embedded within existing clinical workflows. This allows clinicians to use the models without altering their standard procedures, enhancing usability and clinical adoption.

[0046] The present application generates a range of panoramic electroanatomical maps, including activation, propagation, phase, and rotor density maps. These provide a richer and more detailed representation of the heart’s electrophysiological state, supporting better diagnostics and therapeutic planning.

[0047] The present application discloses dynamically enriching 3D heart models with invasive electrophysiological data gathered during catheter procedures, such as electrograms and impedance measurements. This allows for real-time model updates, supporting informed clinical decision-making during procedures.

[0048] The present application is deployed on a web-based platform, enabling real-time access to heart model analyses from various locations. This facilitates remote collaboration and broader clinical engagement, enhancing interdisciplinary diagnostics and treatment planning.

[0049] The present application computes the relationship between cardiac electrical sources and body-surface ECG signals using lead field approach combined with the finite element method. This enables precise non-invasive localization of electrical activity within the heart, which supports more accurate diagnostics and treatment targeting.

[0050] The present application achieves enhanced personalization accuracy by segmenting cardiac structures directly from patient-specific imaging data, such as CT, MRI, or ultrasound. This anatomical specificity supports a more individualized modeling approach.

[0051] The present application incorporates deep learning techniques to automate the segmentation of cardiac structures, thereby improving processing efficiency and potentially reducing variability and manual workload in data interpretation.

[0052] The present application supports the generation of detailed and diverse functional maps, such as rotor density distributions. These more granular visualizations provide enriched information, which may aid in more refined and interpretable clinical diagnostics.

[0053] The present application includes the capability to update and refine models intra- procedurally using real-time electrophysiological data from catheter- based measurements. This real-time feedback loop supports higher targeting accuracy during interventions.

[0054] The present application is designed for integration with existing clinical workflows and diagnostic systems. This compatibility allows for a more streamlined implementation and may reduce the need for procedural changes or additional hardware.

[0055] The present application enables web-based access to its functionality, facilitating collaboration among healthcare professionals across locations. This feature supports modern telemedicine practices and distributed clinical decision-making.

[0056] The present application utilizes a lead field approach combined with the finite element method framework to derive cardiac source activity from surface ECG data with high mathematical precision. This methodological rigor improves the accuracy of inverse electrophysiological reconstructions. Embodiments of the invention are associated with various advantages and / or technical effects.

[0057] There is disclosed a computer-implemented method for generating a three-dimensional activation map of cardiac tissue based on a 12-lead electrocardiogram (ECG). The method comprises receiving 12-lead ECG data of a patient, selecting at least one cardiac cycle from the ECG data corresponding to an arrhythmic morphology, estimating activation times across cardiac tissue based on the selected cardiac cycle, generating a correlation or any other similarity metric between the selected ECG data and precomputed simulated 12-lead ECG signals stored in a database, wherein each simulated 12-lead ECG signal is associated with a simulated activation map, selecting at least one simulated 12-lead ECG signal based on the correlation metric and retrieving the associated activation map(s), generating an initial activation map by averaging the activation maps of the selected ECG signals, generating a final activation map that best fits the measured 12-lead ECG data by optimizing one or more parameters defining the activation map, the one or more parameters including at least one of regional conduction velocity, fiber orientation, and cardiac anisotropy, and displaying the final activation map using a 3D anatomical representation of the heart.

[0058] This application addresses the challenge of improving cardiac diagnostic analysis by integrating advanced algorithms and multi-modality imaging data into existing clinical workflows in a non-disruptive manner.

[0059] It enhances both accuracy and efficiency by employing advanced algorithms to calculate the propagation of electrical excitation across the heart, based on standard 12-lead ECG data. Personalized 3D models of the heart's excitation propagation are constructed using this calculated data, further enriched with CT, MRI, and / or ultrasound imaging inputs. These models are analyzed in real time, allowing healthcare professionals to access the resulting insights seamlessly within their current workflow. Panoramic electroanatom ical maps of the heart are generated using analytical methods.

[0060] The application also integrates invasive electrophysiological data during cardiac procedures, further refining the personalized 3D heart models. By doing so, it meets the demand for a comprehensive and integrated approach to cardiac diagnostics, ultimately contributing to improved patient care and outcomes.

[0061] Using clinical data from surface ECGs, the results are visualized and evaluated in map form. The goal is to generate patient-specific models that closely reflect the individual surface ECG episode.

[0062] A mathematical solution is employed, which may be customized to account for individual patient conditions. The application enables panoramic electroanatomical mapping based solely on standard 12-lead ECG data, thereby avoiding the complexity and cost associated with multimodal imaging, while still providing a viable input modality for full 3D cardiac activation analysis.

[0063] Dynamic, patient-specific 3D heart models are created by integrating ECG-derived excitation data with multimodal imaging sources. This delivers a model that is more physiologically relevant.

[0064] The application supports real-time analysis without necessitating changes to existing clinical workflows, thereby facilitating seamless integration and minimizing training requirements.

[0065] Additionally, non-invasive models are enhanced through the selective inclusion of invasive electrophysiological data, resulting in higher diagnostic fidelity and improved treatment planning.

[0066] In a development, the method further comprises projecting the initial activation map onto a patient-specific mesh by applying anatomical coordinate mapping between the average heart model and a patient-specific anatomical heart model.

[0067] There is further disclosed a computer-implemented method for generating a cardiac activation map. The method comprises receiving data for at least one cardiac cycle obtained from ECG data of a patient, determining at least one parameter representing cardiac activation based on the received data, obtaining from a database at least one activation map, wherein the obtained at least one activation map is selected from the database based on the determined at least one parameter, generating an initial activation map based on the at least one activation map obtained from the database, and generating a final activation map by optimizing the initial activation map.

[0068] This application aims to improve cardiac diagnostic analysis by embedding advanced algorithms and multi-modality imaging data into current clinical workflows, without disrupting established procedures.

[0069] It enhances the accuracy and efficiency of cardiac diagnostics through the use of advanced algorithms that compute the propagation of electrical excitation across the heart from standard 12-lead ECG data. Based on these computations, personalized 3D models of the heart’s excitation dynamics are generated, with the option to supplement these models using CT, MRI, and / or ultrasound data.

[0070] The resulting dynamic, patient-specific 3D heart models deliver a more physiologically representative visualization of cardiac activity. Real-time analysis is supported, enabling healthcare professionals to integrate this information seamlessly into their ongoing clinical routines. This ensures smooth implementation and minimizes the need for additional training.

[0071] Panoramic electroanatomical maps of the heart are calculated using analytical methods. Importantly, the application allows for the selective incorporation of invasive electrophysiological data during cardiac procedures. This enhances the fidelity of the 3D models, thereby improving diagnostic accuracy and supporting more tailored treatment planning.

[0072] By utilizing standard 12-lead ECG data alone, the application facilitates panoramic electroanatomical mapping, offering a cost-efficient, widely available alternative to more complex multimodal imaging systems.

[0073] This approach aims to create patient-specific models that closely replicate the surface ECG episodes observed. It employs a mathematical solution that is adaptable to individual patient conditions, thus supporting a comprehensive, personalized, and integrated strategy in cardiac diagnostics and ultimately contributing to improved patient outcomes.

[0074] In a development, the method further comprises receiving ECG data of the patient.

[0075] In a development, the the ECG data is 12-lead ECG data.

[0076] In a development, the method further comprises selecting the at least one cardiac cycle from the ECG data.

[0077] In a development, the at least one cardiac cycle is selected corresponding to an arrhythmic morphology.

[0078] In a development, the at least one parameter representing cardiac activation is activation times across cardiac tissue.

[0079] In a development, obtaining the at least one activation map from the database comprises comparing the selected at least one cardiac cycle to ECG data stored in the database.

[0080] In a development, the ECG data stored in the database comprises precomputed simulated ECG signals.

[0081] In a development, the step of comparing comprises generating a correlation metric between the selected at least one cardiac cycle and at least one of the precomputed simulated ECG signals.

[0082] In a development, the step of comparing further comprises selecting at least one precomputed simulated ECG signal based on the correlation metric and obtain the at least one activation map which is associated with the selected at least one precomputed simulated ECG signal.

[0083] In a development, generating the initial activation map comprises averaging the at least one obtained activation maps. In a development, generating the final activation map comprises optimizing one or more parameters of the initial activation map.

[0084] In a development, the one or more parameters include at least one of conduction velocity, fiber orientation, and cardiac anisotropy.

[0085] In a development, the method further comprises outputting the activation map to a user.

[0086] In a development, the method further comprises projecting the initial activation map onto a patient-specific model of the heart.

[0087] There is further disclosed a system for generating a cardiac activation map. The system comprises an input interface configured to receive data for at least one cardiac cycle obtained from ECG data of a patient, a database configured for storing at least one activation map, and a processor. The processor is configured to determine at least one parameter representing cardiac activation based on the received data, obtain from the database at least one activation map, wherein the obtained at least one activation map is selected from the database based on the determined at least one parameter, generate an initial activation map based on the at least one activation map obtained from the database, and generate a final activation map by optimizing the initial activation map.

[0088] In a development, the system further comprises an output interface for outputting the activation map to a user.

[0089] There is further disclosed a computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the above method.

[0090] There is further disclosed a computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the above method.

[0091] There is further disclosed a data processing apparatus, comprising means for carrying out the above method.

[0092] There is further disclosed a method for cardiac diagnostic analysis, comprising: a. receiving standard 12-lead ECG data from a patient; b. constructing personalized 3D models of the heart based on the imaging data including CT, MRI, and / or ultrasound; c. employing advanced algorithms to process the ECG data to calculate the propagation of electrical excitation across the heart; d. providing real-time analysis of the underlying heart conditions; e. integrating this analysis into existing clinical workflows without changing those workflows.

[0093] The method enhances diagnostic accuracy by utilizing advanced algorithms to calculate the electrical activity of the heart, providing a more detailed understanding of cardiac function.

[0094] By constructing personalized 3D heart models, the method allows for patient-specific analysis, which can lead to more tailored and effective treatment plans.

[0095] Integration into existing clinical workflows without the need for changes facilitates seamless adoption of the technology, minimizing disruption to healthcare providers and institutions.

[0096] The method provides the advantage of conducting cardiac diagnostic analysis using standard 12-lead ECG data, which is a widely available and commonly used diagnostic tool.

[0097] The method offers the advantage of employing advanced algorithms, including deep learning networks, for processing ECG data and segmenting cardiac structures from multi-modality imaging data. This enhances the accuracy and efficiency of the diagnostic analysis.

[0098] The method provides the advantage of constructing personalized 3D models of the heart based on calculated excitation and multi-modality imaging data. These personalized models allow for a more comprehensive and detailed analysis of the heart's structure and function. The method offers the advantage of generating panoramic electroanatom ical maps of the heart using personalized 3D models. These maps, which include activation, propagation, phase, and rotor density maps, provide valuable insights into the electrical activity and abnormalities within the heart.

[0099] The method provides the advantage of integrating the real-time analysis of the heart models into existing clinical workflows without requiring any changes to those workflows. This ensures seamless integration of the diagnostic analysis into the healthcare professionals' routine practices.

[0100] In a development, the method further comprises advanced algorithms including deep learning networks for segmentation of cardiac structures from the multi-modality imaging data.

[0101] The inclusion of deep learning networks for segmentation speeds up the process and improves the precision of identifying cardiac structures, leading to more accurate 3D heart models.

[0102] The use of deep learning enhances the method's ability to adapt and improve over time as more data is processed, potentially increasing the reliability of diagnostic outcomes.

[0103] By automating the segmentation process, the method reduces the potential for human error and decreases the time required for data processing.

[0104] In a development, the method further comprises a segmentation of cardiac structures being performed automatically by the deep learning networks.

[0105] Automatic segmentation by deep learning networks streamlines the diagnostic process, allowing for quicker generation of personalized heart models.

[0106] The automation of segmentation tasks frees up medical professionals to focus on more critical aspects of patient care and analysis. Consistency in segmentation is improved with the use of standardized deep learning protocols, reducing variability between different operators or institutions.

[0107] In a development, the method further comprises generating panoramic electroanatom ical maps of the heart using the personalized 3D models.

[0108] Generating panoramic electroanatom ical maps provides a comprehensive view of the heart's electrical activity, aiding in the identification of arrhythmias and other abnormalities.

[0109] The panoramic maps enhance the ability to visualize and understand complex cardiac phenomena, which may lead to more effective interventions.

[0110] The use of personalized 3D models to create these maps ensures that the electroanatom ical data is highly relevant to the individual patient, potentially improving the outcomes of diagnostic and therapeutic procedures.

[0111] In a development, the method further comprises panoramic electroanatom ical maps including at least one of activation maps, propagation maps, phase maps, and rotor density maps.

[0112] Including various types of electroanatomical maps, such as activation, propagation, phase, and rotor density maps, offers a multifaceted assessment of cardiac electrical function.

[0113] The availability of diverse map types allows clinicians to select the most appropriate visualization for the specific diagnostic challenge at hand, enhancing clinical decisionmaking.

[0114] The detailed insights provided by these maps can facilitate the identification of the precise locations of cardiac dysfunctions, which enables planning targeted treatments such as ablation therapy. In a development, the method further comprises augmenting the personalized 3D heart models with invasive electrophysiological data during a cardiac procedure.

[0115] Enhancing the personalized 3D heart models with invasive electrophysiological data provides a more accurate representation of the patient's cardiac electrophysiology, leading to improved diagnostic precision and tailored therapeutic strategies.

[0116] The integration of real-time electrophysiological data during cardiac procedures allows for immediate adjustments to treatment plans, potentially reducing procedure time and improving patient outcomes.

[0117] In a development, the method further comprises invasive electrophysiological data including at least one of electrograms, impedance measurements, catheter positions, created meshes, and maps.

[0118] Incorporating a comprehensive set of invasive electrophysiological data, such as electrograms and impedance measurements, enables a detailed analysis of cardiac electrical activity, which can facilitate the identification of arrhythmogenic substrates and guide ablation therapy.

[0119] The use of created meshes and maps in conjunction with catheter positions enhances the spatial resolution of the electrophysiological assessment, allowing for precise localization of pathological areas within the heart.

[0120] In a development, the method further comprises utilizing a web-based platform to provide access to the real-time analysis of the heart models for healthcare professionals.

[0121] A web-based platform for accessing real-time analysis of heart models ensures that healthcare professionals can collaborate and make informed decisions from remote locations, enhancing the accessibility and efficiency of cardiac care.

[0122] The platform's ability to provide real-time data supports dynamic decision-making during cardiac procedures, potentially reducing the risk of complications and improving the overall safety of the intervention. In a development, the method further comprises multi-modality imaging data being processed to identify areas of fibrotic tissue and / or scarring.

[0123] Processing multi-modality imaging data to identify fibrotic tissue and scarring enables the detection of structural abnormalities that are often associated with arrhythmias, thereby aiding in the risk stratification and management of patients with cardiac diseases.

[0124] The identification of areas of fibrosis and scarring can inform the selection of ablation targets during catheter ablation procedures, potentially increasing the success rate and reducing the likelihood of arrhythmia recurrence.

[0125] In a development, the method further comprises employing a finite element solver to establish a relationship between cardiac electrical sources and ECGs measured on the body surface.

[0126] Employing a finite element solver to establish the relationship between cardiac electrical sources and body surface ECGs allows for non-invasive mapping of cardiac electrophysiological events, which can reduce the need for invasive diagnostic procedures.

[0127] The use of a finite element solver enhances the computational efficiency of solving the inverse problem of electrocardiography, enabling faster processing times and the potential for real-time clinical application during cardiac procedures.

[0128] There is further disclosed a method for cardiac diagnostic analysis (the analysis of the heart models and / or the ECGs), comprising: a. ECG data that is acquired from a patient using a standard 12-lead ECG machine; b. constructing models that are personalized to a patient and that may also be personalized to a specific cardiac procedure based on the propagation of electrical excitation across the heart, provided by multi-modality imaging data such as CT, MRI, and / or any type of ultrasound, including, but not limited to, doppler ultrasound, color doppler ultrasound, and / or contrast-enhanced ultrasound; c. employing algorithms that are advanced enough in their field of application to calculate a more sophisticated analysis to process the standard 12-lead ECG data to calculate the propagation of electrical signals across the cardiac model; d. providing an analysis of cardiac models in a near-real time without requiring changes in clinical workflows of the heart models or a collection of heart models that include cardiac structures and / or electrical signals; e. integrating this process that can be performed in parallel with existing clinical workflows without altering the flow of data from one patient to another and without necessitating any changes that may occur in existing clinical workflows without causing significant disruptions or delays to those workflows and without significant changes, such as the creation of new workflows.

[0129] In a development, the method further comprises a neural network that utilizes artificial intelligence to process cardiac structures from imaging data of cardiac structures. The term segmentation of cardiac structures is used herein to refer to a process for automatically processing cardiac structures from imaging data, such as CT, MRI, and ultrasound from the multi-modality imaging data.

[0130] In a development, the method further comprises generating panoramic electroanatom ical maps in the form of a series of at least one image of the heart, which may be taken in real-time or virtual time, including at least one of activation maps which represent the phase of the electrical excitation, propagation maps which represent the propagation of electrical excitation across the heart, phase maps, and rotor density maps of the heart.

[0131] In a development, the method further comprises augmenting the personalized heart models with invasive electrophysiological data during a procedure that is performed on a patient's heart, such as a cardiac ablation procedure, a cardiac resynchronization therapy procedure, a cardiac pacemaker procedure, a cardiac defibrillation procedure, a cardiac pacing procedure, a cardiac ablation procedure, or a cardiac resynchronization therapy procedure.

[0132] In a development, the method further comprises utilizing invasive electrophysiological data in the form of cardiac excitation and / or imaging data, including at least one of the electrical potentials measured on the body surface, the electrograms of the heart, the positions of at least one catheter in the heart, meshes that are created by an invasive system, and maps in the form of a collection of at least one image of the heart, including panoramic electroanatom ical maps.

[0133] In a development, the method further comprises utilizing a platform that provides access to the patient’s analysis in real-time, with the ability of a web-based platform, such as a website, to gain access to clinical data to the real-time analysis of the heart models for healthcare professionals.

[0134] In a development, the method further comprises multi-modality imaging data being processed to identify areas of cardiac structures that may be affected by electrical excitation, fibrotic tissue and / or scarring.

[0135] In a development, the method further comprises employing a finite element solver in the form of a bilateral-element solver based on a mathematical model that uses finite elements to establish a relationship between cardiac electrical sources and ECGs measured on the body surface. A relationship between cardiac electrical sources and the standard 12-lead ECGs measured on the body surface is then established.

[0136] The 3D Modeling Software according to the application comprises several modules.

[0137] A segmentation module takes DICOM raw data from the dedicated cardiac imaging obtained with CT 12 and / or MRI 11. This data can be displayed in various views and number of windows. E.g. 3 standard planes, while a 4th window is dedicated for the rendered volume.

[0138] Trained deep-learning networks find the bounding box around the cardiac area and classify all voxels as belonging or not belonging to the cardiac structure of interest. The anatomical structure to be segmented can also be selected from the drop-menu when loading DICOM data. The segmentation module reconstructs the 3D voxel of the cardiac chambers (atria or ventricles), extracts the surface mesh and pre-processes it making it suitable for further numerical calculations.

[0139] 12-lead ECG data is loaded. Time-series for any number of selected leads can be visualized. There are multiple visualization options including scaling, zooming, scrolling, interval selection. Peaks are found based on the wavelet decomposition of the whole ECG signal. The templates found are then used for reliable peaks' thresholding. A deep learning-based ECG analysis is used for rhythm classification and abnormality detection based on the open datasets and our own proprietary data. Fully automatic Al-based ECG delineation is applied for further analysis.

[0140] Invasive data: off-line and / or live data from an invasive system can be loaded in realtime or after the procedure into the 3D modeling software. These data usually include electrograms, impedance measurements, catheter positions, created meshes and maps. The EP signals from an invasive system are always local in nature. The 3D modeling software takes these local snapshots and incorporates them into the solution scheme to increase the local accuracy and fine-tune the model to improve the global accuracy as well.

[0141] Maps module 25: Given the segmented patient-specific heart or any other appropriate 3D heart model and selected ECG fragment, a 3D panoramic electroanatomical map can be created. The maps include activation map, propagation map, phase maps, rotor density maps. By selecting different time intervals one can re-compute or compute new maps, save them individually or as the whole project.

[0142] The application provides a comprehensive software system designed for enhancing cardiac therapy planning and outcome monitoring. If available and / or required it integrates and analyzes multi-modality medical imaging data to create personalized, detailed 3D heart models. These models support precise diagnostics, therapy planning, and the monitoring of therapeutic outcomes, aiming to improve patient-specific treatment strategies and outcomes in cardiac care.

[0143] Unlike existing methods that require complex electrode setups, this system can work with textbook-standard heart models, standard 12-lead-ECG configurations and provide real-time analysis.

[0144] The goal of the invention is a software system that simplifies and enhances cardiac diagnostic processes through non-invasive methods. It uses standard ECG data and advanced algorithms to construct personalized 3D electroanatomical models for therapy planning. It integrates various data types, without altering clinical procedures, and it offers insights for decision-making rather than prescribing specific interventions. The application works with standard ECG and possibly textbook heart models, streamlining clinical processes without requiring new ones. The 3D modeling software does not aim to replace medical expertise but rather to enhance it, providing visualization and analytical tools to aid decision-making without making clinical decisions itself. This respects the conservative nature of medical adoption and focuses on assisting rather than automating clinical judgments. The software system according to the application makes it easier and better to diagnose heart problems. It takes normal 12-lead-ECG data (a test that checks how your heart is doing) and uses smart algorithms to create personalized 3D models of the excitation spreading in the heart. Unlike other methods that need complicated setups, this system also works with standard - i.e. pre-defined and stored in a database - heart models and can analyze data in real time. It combines different kinds of data, like detailed heart scans (CT, MRI, ultrasound) and other data, without changing the usual clinical procedures.

[0145] The application provides a software that processes multi-modality data and merges them to create a personalized 3D electroanatomical map for enhanced treatment planning for cardiac arrhythmias' patients.

[0146] The multi-modality data include, but not limited to: a) Imaging, e.g. computerized tomography (CT) and I or magnetic resonance imaging (MRI), b) functional imaging, e.g. MRI or positron emission tomography (PET), c) non-invasive electrocardiogram (ECG), d) invasive electrophysiological data, e.g. intracardiac electrical signals and maps.

[0147] These types of data provide structural and functional cardiac information for assessment of the underlying heart condition. Imaging allows creating patient-specific anatomical models of the heart together with delineation of structurally abnormal areas, such as fibrotic tissue and I or scarring (due to e.g. infarction). Highly-individualized cardiac anatomy influences generation of electric signals arriving and measured at the human body surface as electrocardiogram (ECG).

[0148] For processing the imaging data, segmentation is performed in a two-step process. First, the location of the heart is identified within the provided series of torso scans. Secondly, the heart itself is segmented from the bounding box resulting from the first step. For these purposes, two deep neural networks are applied.

[0149] Dedicated deep-learning U-Net based networks are used. The final solution consists of two networks (plainly speaking, cutting out the region of interest and then segmenting fine structures), whose architectural parameters were fine-tuned for the data available. Architectural parameters are: number of layers, number of convolution kernels, dropout ratios, and connectivity between the layers.

[0150] For processing the map creation step, the relation between cardiac electrical sources and ECGs measured on the body surface is provided by the Poisson equation, which is solved with the dedicated finite element solver. The volume conductor is assumed to be homogeneous, however any number of inhomogeneities can be incorporated into the model and available data from functional imaging. As a result, a non-linear relationship is established between activation time on the cardiac surface and the ECG at the respective electrode positions. This non-linear equation is solved with the dedicated optimization routine.

[0151] For detection of rotational activities and calculation of the rotor density maps the solution is stabilized by interpolating signals in between the electrodes, then reconstruct electrograms, make a Hilbert phase transform and analyze the stability of the core rotor drives.

[0152] Generating the output comprises the following steps: a) distinct panoramic maps showing physiological scalar (e.g. activation times) or vector (e.g. propagation) maps b) local arrhythmogenic sources c) ECG characteristics (e.g. duration of waves, fractionation indices) d) anatomical metrics (e.g. left atria volume) e) risk stratification and therapy response prediction.

[0153] ECG together with the anatomical information provide a panoramic view of the electrical activity of the heart. However, it sometimes requires high local resolution, which could be only delivered by invasive mapping. Integrating invasively obtained signals, the new software generates the first-in-class methodology of merging anatomical I structural data together with all-chambers panoramic view of the electrical cardiac activity and local electrophysiological tissue characteristics for fine-tuning computational models of the heart. This fine-tuning allows a truly personalized approach for planning and therapy in patients with adverse cardiac events.

[0154] The application delivers a holistic view on the cardiac state of condition and can be used for improved guidance and therefore success rates of ablation therapies, both acutely and long-term.

[0155] The application takes into account that

[0156] 1) the electrical signals coming from the heart to the body surface are affected by patient-specific heart anatomy and position, and that

[0157] 2) intracardiac interventions also do require the anatomy of cardiac chambers to map out the most probable excitation pathways and, based on this knowledge, deliver targeted ablation pulses to the affected areas, while preserving all healthy tissue.

[0158] The application incorporates chambers patient-specific geometries into EP routine by

[0159] 1) lowering logistic hurdles arising from scheduling extra CT or MR protocols

[0160] 2) automatic segmentation of cardiac structures.

[0161] According to the application, ECG is used for providing a global panoramic picture of electrical excitation in the heart, restricting analysis to the local area of catheter manipulations. Mapping of the whole chamber is performed in a consecutive way by mapping out local zones of potential interest.

[0162] The software solution of the application integrates patient-specific anatomy and ECG information into the calculations and augments the invasive signals, thus providing a more complete view of the cardiac electrical activity refined by locally recorded electrograms.

[0163] The device of the application provides:

[0164] 1) fully automatic segmentation of cardiac chambers incorporated to the EP procedural logistics and planning; 2.1) non-invasive 12-lead ECG interpretation without other signals;

[0165] 2.2) non-invasive 12-lead ECG interpretation in the context of available invasive signals

[0166] 3) augmentation of invasive maps by integrating the two previous advancements.

[0167] The device of the application incorporates and integrates different sources of data essential for successful performance of cardiac EP procedures. These data include but not limited to prerecorded and running ECG acquisitions, export from imaging modalities e.g. CR or MRI, direct input from an invasive EP system used for mapping and ablation.

[0168] Such an export can be done either manually or automatically, which is already implemented within some invasive systems. It can be furthermore enhanced by clinical test data including blood analysis, overview of patient-specific gene expressions etc. Such biomarkers have been shown to have potential in risk predisposition and treatment outcome prediction in some cohorts of patients. This could be achieved by collecting all-round patient information concerning distinct clinical tests and analysis and utilizing it within a framework of an artificial intelligence system tailored for a specific diagnosis or outcome prediction task.

[0169] All pieces of available modalities are inter-connected with the highest priority given to the invasive EP data being processed in real (or near real) time. Non-invasive 12-lead ECG augments the information by creating a more reliable panoramic picture of the whole heart electrical excitation patterns. This augmentation is done by the dedicated computer routine.

[0170] Upon availability of more precise anatomy, the computational model becomes realistic in terms of representing electrical sources and providing operating physicians with the most complete view on the patients' heart. The final result of computerized heart models include but are not limited to a) distinct panoramic maps showing physiological scalar (e.g. activation times) or vector (e.g. propagation) maps b) local arrhythmogenic sources c) ECG characteristics (e.g. duration of waves, fractionation indices) d) anatomical metrics (e.g. left atria volume) e) risk stratification and therapy response prediction.

[0171] Instead of CT / MRI-data which provide patient-specific 3D-models of the heart, also general 3D-heart models can be used. Although not patient-specific results for these kind of models still provide valuable information for the physician with respect to nature and source of arrhythmia before the intervention. During intervention patient-specifity is achieved with local catheters as a probe.

[0172] In a similar manner, incomplete 3D-models of the heart, e.g. from ultrasound can be completed by respective algorithms.

[0173] By adding respective interfaces (e.g. fibrosis data from MRI) further data can be incorporated in the analysis of 3D-excitation spreading in the heart.

[0174] The 3D modeling software according to the application comprises the following software modules: segmentation module 20

[0175] Automated analysis of ECG 21

[0176] Segmentation of cardiac anatomy

[0177] 3D heart reconstruction based on CT, MRI, ultrasound or other sources electroanatom ical map 25

[0178] Automated reconstruction of 3D maps based on standard tools and processes web-based platform

[0179] Web-based real-time data platform for healthcare professionals

[0180] All modules run locally on standard computer hardware or browser-based on a webplatform. The segmentation module provides software interfaces for respective data sources as described before.

[0181] The 3D modeling software fills the crucial niches hindering adoption of patient-centric cost-effective treatment strategies. The provided upgrade is ensured by

[0182] 1) the approach based on standard processes and tools, thus not requiring additional logistical efforts critical in the clinical environment

[0183] 2) automated evaluation of all components with the insights for all involved healthcare units - resident general practitioners, cardiologists, hospitals, telemedicine centers, insurance companies

[0184] 3) easy web-access to all pieces of complementary information and effective patientcentric communication between the aforementioned actors, e.g. ECG analysis from a GP's visit is forwarded and integrated into EP treatment planning module

[0185] 4) time and cost reduction together with lower readmission numbers, allowing better services to the whole population

[0186] 5) generally, a more holistic approach to cardiovascular health via digital advances.

[0187] The software of the application allows the better use of existing hardware for cardiac arrhythmia diagnosis - 3-dimensional electro-anatomical maps of the heart - by interfacing with existing hardware and automatically integrating with standard non- invasive processes, significantly improving patient outcomes and healthcare provider efficiency.

[0188] The application creates 3-dimensional electro-anatomical maps of the heart based on conventional 12-lead ECG recordings, by simply interfacing to existing ECG-hardware. These maps can be used in the diagnosis of cardiac arrhythmias such as atrial fibrillation. Until now, these maps could only be created invasively during a catheter procedure or using complex electrode vests with around 200 electrodes.

[0189] The results are made available to everyone involved in the healing process simply and in real time via a web platform. This significantly improves the patient's therapy and prognosis and makes the healing process more efficient. This and the simplified data exchange and generation of patient-specific data result in significant cost savings.

[0190] An example implementation of a method according to the invention is described below. The method can comprise the steps patient data acquisition and upload into the system, map generation, and visualization and analysis. Patient data is acquired and uploaded into the system as described in the following. A digital 12-lead ECG is obtained during routine clinical evaluation or prior to an electrophysiological intervention. The ECG is recorded in accordance with applicable clinical standards and exported in a digital format compatible with the system (e.g., XML, DICOM).

[0191] The ECG file is uploaded to a mapping system interface wherein a user (e.g., a physician or technician) selects one or more cardiac cycles exhibiting the arrhythmic morphology of interest. Selection may be performed manually or semi-automatically via built-in tools that facilitate beat identification). Table 1 below shows an example of the numerical data of 12 standard-channels for such an interval at 1 kHz sampling rate

[0192] (values are provided in microvolts).

[0193] Tab e 1: example of numerical data of 12 standard-channels at 1kHz sampling rate

[0194] When an ECG fragment becomes available, the map generation can be done in two ways: a) based on the personalized or b) using a default, or pre-defined average, anatomical model of the heart. For a), the imaging data from CT, MRI or ultrasound are segmented and geometrical models of the human torso and heart, and respective meshes for numerical calculations are created. If more advanced imaging protocols, e.g. late Gadolinium enhancement MRI, are available for the patient, cardiac tissue parameterization can be performed in order to identify areas of fibrosis and scarring. This information can be integrated into the available geometrical mesh by assigning unique tissue properties to the discretization elements of the mesh (triangles, tetrahedra, hexahedra etc.). After the computations, this mesh is also used for visualizing maps for selected ECG fragments.

[0195] In case personal imaging data are not available and, therefore, it is not feasible to derive a personalized geometrical model of the human body, a default model with the properties described above can be used for calculations.

[0196] A map is generated as described below, after patient data was acquired and uploaded into the system.

[0197] For the creation of 3D electro-anatomical maps from conventional recorded 12-lead- ECGs a database of calculated cardiac activation models and simulated 12-lead-ECGs is created. Creating the database comprises the steps of modeling a cardiac activation model and of simulating numerical 12-lead ECG.

[0198] The cardiac electrical activation is modeled using the anisotropic eikonal equation, which simulates the propagation of electrical waves through the myocardium by accounting for anisotropic conduction velocities inherent in cardiac tissue. This equation has activation times (AT) as unknowns and describes how the activation propagates through the tissue, with the conduction velocity tensor capturing the anisotropic nature of myocardial fibers leading to faster excitation spread along the fiber direction compared to across it.

[0199] To solve this equation on a tetrahedral domain that represents the complex cardiac geometry, a fast iterative method is implemented as described in Fu Z, Kirby RM, Whitaker RT. A FAST ITERATIVE METHOD FOR SOLVING THE EIKONAL EQUATION ON TETRAHEDRAL DOMAINS. SIAM J Sci Comput. 2013;35(5):c473- c494. doi: 10.1137 / 120881956. PMID: 25221418; PMCID: PMC4162315. In this approach, the myocardium is discretized into a tetrahedral mesh, and the AT at the mesh nodes are iteratively updated until the convergence is achieved.

[0200] Numerical 12-lead ECG is further simulated. Cardiac transmembrane voltages (TMV) are calculated from the AT using a cellular action potential template. The TMV gradients calculated within all tetrahedra in the cardiac mesh are then equivalent to the current dipole source distribution. For 12-lead ECG simulation, the lead field approach is utilized as described in Potse M. Scalable and Accurate ECG Simulation for Reaction-Diffusion Models of the Human Heart. Front Physiol. 2018 Apr 20;9:370. doi: 10.3389 / fphys.2018.00370. PMID: 29731720; PMCID: PMC5920200. In this method, the electrical activity of the heart is modeled as a distribution of current dipoles that generate a potential field throughout the torso. The potentials measured by each lead are obtained by integrating the contributions of the cardiac electrical sources weighted by the corresponding lead field. The lead field for each electrode is derived by solving a corresponding Laplace equation within the torso domain, taking into account the geometry and tissue conductivities. The solution to these equation provide spatial maps describing how potentials generated within the myocardium are projected to the electrode locations on the body surface.

[0201] Integration of different source distributions result in 12-lead ECG signals varying in both amplitude and morphology. Thus, by incorporating accurate lead fields computation and cardiac activation models, this simulation approach produces 12-lead ECG signals that reflect spatio-temporal heterogeneity of the cardiac electrical activity.

[0202] A simulated dataset containing realistic cardiac electrical activation patterns on a three- dimensional heart / torso domain are further generated. This dataset is constructed by varying key parameters related to the activation source and conduction velocity tensors across the myocardium. Main goal of this step is to simulate a range of physiological and pathophysiological conditions by varying the location and strength of activation sources and by modulating conduction velocities.

[0203] For the activation source, e.g., focal excitation is considered, whereby a single point within the myocardium is chosen to serve as the activation trigger. This method allows for simulation of focal activation patterns, effectively mimicking the electrical behavior observed under arrhythmic events.

[0204] Modulation of conduction velocity is another aspect of the dataset creation. The model incorporates anisotropic conduction by varying the conduction velocities in the directions along and transversal to the myocardial fiber orientation. This allows the simulations to capture the physiological effect of faster conduction along the fiber and slower conduction across it. To further enrich the simulated dataset, pathological conditions were simulated by introducing regions with reduced conduction velocity that mimic scar tissue or fibrotic areas. The simulation process involves solution of the forward problem, where the cardiac electrical activation model, based on the anisotropic eikonal equation, was used to compute activation times across the tetrahedral mesh representing the heart. Multiple simulations are run while systematically varying both the activation source and conduction velocity parameters using Halton scheme of quasi-Monte Carlo method. The resulting activation maps, along with the detailed parameter settings such as source locations and conduction velocities, are stored as part of the dataset. Moreover, synthetic (simulated) 12-lead ECG signals were generated using the lead field approach.

[0205] The simulation outputs are (as elements of the database) activation Maps, parameter metadata, and simulated ECG signals. The activation maps provide high-resolution datasets containing AT across the myocardium under varied conditions. The parameter metadata provide comprehensive log of the parameter variations for each dataset instance, including details on activation source locations, onset times, and conduction velocity values. The simulated ECG signals provide corresponding 12-lead ECG signals generated via the lead field method.

[0206] An algorithm for the map generation is described below and is further described in Fig. 10 and Fig. 11 of the present application.

[0207] Upon selection of the target interval the system initiates computational solution:

[0208] - In instances where patient-specific anatomical imaging (e.g., MRI, CT, ultrasound) is available a personalized three-dimensional anatomical mesh is generated and utilized for the simulation (see Fig. 11).

[0209] - In the absence of such imaging a pre-defined anatomical model is employed (see Fig. 10). The selected ECG fragment is processed to yield estimated activation times from which a three-dimensional activation map is computed.

[0210] An inverse mapping algorithm is used to identify the best-matching activation maps from the set of simulated ECG signals. The process leverages signal correlation to compare simulated ECG data with the clinical patient’s ECG recording. Specifically, the algorithm calculates the correlation between the patient’s 12-lead ECG and each simulated 12- lead ECG in the dataset. The top three simulated signals exhibiting the highest correlation are selected, and their corresponding activation maps are retrieved as candidate solutions. First, the simulated dataset, including activation maps and their associated ECG signals generated via the lead field approach, is organized for efficient comparison. Second, the real patient ECG is pre-processed in the same way as the simulated data to ensure consistency in signal characteristics.

[0211] Prior to performing the correlation analysis, a resampling step is applied to the simulated ECG signals to accurately match the QRS complex length. This resampling adjusts the temporal resolution of the simulated ECGs using fast Fourier transform resampling to ensure that the duration and morphology of the QRS complex are properly represented. This normalization step prevents variations in sampling frequency or resolution from distorting the temporal features of the QRS complex, which provide for accurate correlation with the patient’s ECG.

[0212] Next, a correlation analysis is performed. For each simulated ECG signal, the Spearman correlation coefficient is computed with the real patient’s ECG. By ranking the simulated signals based on their correlation scores, the algorithm identifies the top three signals that most closely resemble the patient’s ECG.

[0213] In the final step, the activation maps corresponding to these top three signals are selected and averaged to produce an activation map (‘initial map’ in the following), to be used by its own and be subject for further optimization. This averaging helps to reduce noise and enhance the robustness of the estimated activation pattern. The final solution step is the optimization of the activation related parameters to match the calculated and measured ECG.

[0214] In the case of using the average heart model, the initial map is taken as an initialization for the optimization routine. In case a patient-specific anatomy of the heart is available, i.e. the CT / MRI data were segmented and a discretized mesh was created for calculations, the initial map is projected onto the patient-specific geometry (mesh). For such a projection universal ventricular and atrial coordinates are used to anatomically parametrize the patient-specific mesh (Pankewitz LR, Hustad KG, Govil S, Perry JC, Hegde S, Tang R, Omens JH, Young AA, McCulloch AD, Arevalo HJ. A universal biventricular coordinate system incorporating valve annuli: Validation in congenital heart disease. Med Image Anal. 2024 Apr;93:103091. doi: 10.1016 / j. media.2024.103091. Epub 2024 Jan 19. PMID: 38301348; PMCID: PMC11227738. Roney CH, Pashaei A, Meo M, Dubois R, Boyle PM, Trayanova NA, Cochet H, Niederer SA, Vigmond EJ. Universal atrial coordinates applied to visualisation, registration and construction of patient specific meshes. Med Image Anal. 2019 Jul;55:65-75. doi:

[0215] 10.1016 / j. media.2019.04.004. Epub 2019 Apr 17. PMID: 31026761 ; PMCID: PMC6543067.). The one-to-one correspondence between the average and personalized heart mesh coordinates is established, and the functional parameter values (including the initial map) are translated from the former to latter model.

[0216] In the final step, non-linear optimization over parameter space, including regionally varying conduction velocity, fiber orientation and cardiac anisotropy ratio is performed to calculate the activation times map producing an ECG matching the 12-lead measurements.

[0217] Visualization and analysis can be performed after map generation. The resulting activation map is displayed via the system’s visualization module which renders isochronal map and propagational patterns superimposed on the anatomical heart model. The user may interactively examine regions of interest.

[0218] BRIEF DESCRIPTION OF DRAWINGS

[0219] The present disclosure is illustrated by way of example and not limited in the accompanying figures in which like reference numerals indicate similar elements. Embodiments of the application will now be described with reference to the attached drawings:

[0220] Figure 1 shows a patient that is connected to a system of the application;

[0221] Figure 2 shows a schematic overview of the 3D modeling software of Fig. 1 .

[0222] Figure 3 shows a flowchart of a method for time course comparison of 3D electroanatomic maps for arrhythmia monitoring.

[0223] Figure 4 shows a treatment sequence of a patient.

[0224] Figure 5 shows a flowchart of a method for generating a cardiac activation map.

[0225] Figure 6 shows a flowchart of a method for noninvasive cardiac mapping.

[0226] Figure 7 shows a flowchart of a method for segmenting cardiac anatomical structures from medical imaging data for use in non-invasive cardiac mapping.

[0227] Figure 8 shows a flowchart of a method for analyzing ECG signals for use in non-invasive cardiac mapping. Figure 9 shows a flowchart of a method for generating a three-dimensional activation map of cardiac tissue based on a 12-lead electrocardiogram (ECG).

[0228] Figure 10 shows a flowchart of a further method for generating a three-dimensional activation map of cardiac tissue based on a 12-lead electrocardiogram (ECG).

[0229] Figure 11 shows a flowchart of a further method for generating a three-dimensional activation map of cardiac tissue based on a 12-lead electrocardiogram (ECG).

[0230] Figure 1 illustrates a comprehensive system designed for cardiac diagnostics and therapy planning. The patient 1 is centrally depicted with various connections to diagnostic equipment. These include an ECG machine 10 directly connected to the patient, displaying ECG data through various leads labeled I, II, III, aVR, aVL, aVF, V1 through V6. The patient 1 is also linked to a magnetic resonance imaging (MRI) device 11 or a computerized tomography (CT) machine 12. These imaging devices are capable of generating detailed anatomical data of the patient's heart.

[0231] Fig. 1 also shows a 3D heart model representation 16 displayed on a computer screen 15, which is part of a data integration hub computer 14. This computer 14 is equipped with a 3D modeling software 17, allowing for sophisticated analysis and modeling of the heart based on the data collected from the MRI 11 and CT 12 devices, along with the ECG data. An ultrasound machine 13 is also depicted that is connected to the patient 1.

[0232] The system is designed to integrate and analyze multi-modality medical data, facilitating enhanced diagnostics and personalized therapy planning for cardiac conditions.

[0233] The figure demonstrates the components and workflow of a method for cardiac diagnostic analysis. At the center, there is a representation of a patient 1 , from whom standard 12- lead ECG data 30 is being collected. Leads labeled with V1 to V6, aVR, aVL, and aVF along with I, II, III indicate the various positions on the patient's body where ECG data is captured. To the left of the patient are three rectangles representing sources of multimodality imaging data: an MRI machine 11 , imaging data that includes CT (Computerized Tomography) 12, and an ultrasound device 13. These data sources are interconnected with the central figure, the patient, suggesting that data from the patient and imaging modalities are collected and processed together. On the right side of the figure, we see a data integration hub computer 14 with a computer screen 15 displaying a personalized 3D model of the patient's heart 16. This implies that the ECG data and imaging data are processed and integrated to construct this heart model. Below the computer is a depiction of 3D modeling software 17, which is likely the application used to generate the heart models from the collected data. The ECG machine 10 is symbolically represented, with its connection to the 3D modeling software, suggesting the software's capability to process the ECG data. The combination of these elements showcases a comprehensive system designed to provide real-time analysis of cardiac health, supporting professionals in the diagnosis and treatment of heart-related conditions.

[0234] Figure 2 depicts the workflow of the 3D modeling software “LlniD” 17, focusing on the integration and analysis of diverse medical data for cardiac care. The diagram features several modules and data pathways, and it shows a visualization of “UniD”'s data flow, from input to analysis and output. It covers the different types of data inputs (DICOM, ECG, invasive data, patient history), how they're processed (segmentation, merging, analysis), and the types of outputs generated (3D models, EP maps, ECG analysis).

[0235] The core of the system is represented by two primary software modules: the segmentation module 20 and the ECG analysis module 21. The segmentation module 20 processes DICOM data to create 3D volumes and meshes of cardiac structures. This involves merging meshes from different imaging sources to enhance the anatomical model of the heart. This module receives inputs from DICOM, electrophysiological (EP) data, and ECG data 30, and is closely linked to the patient 1.

[0236] The ECG analysis module 21 focuses on processing ECG data to identify critical intervals and peaks which are crucial for accurate cardiac analysis. The outputs from this module feed into the “MAPS” module 25, which is tasked with generating panoramic views of the heart's electrical activity.

[0237] The maps module 25 integrates outputs from both the segmentation and ECG analysis modules to produce comprehensive electroanatomical maps. These maps provide detailed visualizations of segmented structures, enhanced panoramic views, and specific cardiac metrics such as peaks, intervals, and risk scores.

[0238] The block diagram is labeled "IINID" at the top, which indicates a system or component name. Inputs on the left side include ECG data 30, patient information 1 , and imaging data in DICOM format. These inputs feed into two primary modules: the segmentation module 20 and the ECG analysis module 21. The segmentation module processes DICOM inputs to create 3D volumes or meshes of the heart, represented by the term "3D VOLUMES / MESHES." This could relate to claim 2 and claim 3, which pertain to advanced algorithms including deep learning networks for segmentation and the automatic performance of the segmentation. Subsequently, these 3D volumes merge with electrophysiological data (EP) to produce electrophysiological maps (EP) indicating local signals. This process aligns with claim 4, which involves generating panoramic electroanatom ical maps using personalized 3D models, and claim 6, which includes augmenting the personalized 3D heart models with invasive electrophysiological data. On the right side, the ECG analysis module 21 processes ECG data to filter and select specific intervals, which then feed into a "MAPS" panoramic module 25. It is not explicitly labeled in the figure but likely corresponds to the maps module mentioned in the claim set. Outputs from the system are categorized by data type on the right side. DICOM data results in "SEGMENTED STRUCTURES," which likely pertain to the created 3D heart models (claim 1). Electro-physiological (EP) inputs lead to an "ENHANCED PANORAMIC VIEW" which could be related to the panoramic electroanatomical maps (claim 5). The ECG analysis provides "PEAKS, INTERVALS, RISK SCORES," capturing elements of the ECG that could be deemed significant for diagnostics and potential risk assessment. This could correlate with claim 10, which mentions employing a finite element solver. Lastly, patient inputs contribute to "OUTCOME PREDICTION TREATMENT I FUTURE PLANNING," suggesting that the system's analysis offers prognostic insight and aids in determining the course of treatment or management for the individual. The figure does not explicitly mention a web-based platform or the integration into existing clinical workflows, but these elements could be implicit, as they are included in claim 8 and claim 1 , respectively. Overall, the diagram encapsulates a holistic approach to cardiac diagnostics using multi-modal data and advanced processing techniques.

[0239] Figure 3 shows a flowchart of a method for time course comparison of 3D electroanatomic maps for arrhythmia monitoring. The method comprises the step of receiving 12-lead ECG data 30 at multiple timepoints such as a set 32 of ECG data comprising a first set 34 of ECG data at a first timepoint and a second set 36 of ECG data at a second timepoint. The method further comprises generating 3D electroanatomic maps such as a first electroanatom ic map 40 representing the first timepoint and a second electroanatomic map 42 representing the second timepoint.

[0240] The method further comprises registering and aligning the generated electroanatomic maps to a common coordinate system and computing activation metrics. The activation metrics can be quantitative metrics such as first quantitative metrics 44 representing the first timepoint and second quantitative metrics 46 representing the second timepoint. The method further comprises comparing the generated first electroanatomic map 40 and the second electroanatomic map 42 with each other and also comparing the first quantitative metrics 44 and the second quantitative metrics 46 with each other.

[0241] The method further comprises detecting recurrence of arrhythmias or new arrhythmias and generating a treatment decision report based on the comparison and the detection of arrhythmias. The method further comprises displaying the report on a user interface.

[0242] Figure 4 shows a treatment sequence of a patient. External therapy is followed by a regular follow and obtaining a standard 12-lead ECG. UniMAP electroanatom ical map is subsequently created and analyzed. A physician then decides on the treatment of the patient based on the analysis of the map. Depending on the decision of the physician, the patient undergoes a further external therapy or the medication of the patient is adjusted. After adjusting the medication or the further external therapy, the sequence starts again with a further regular follow up.

[0243] Figure 5 shows a flowchart of a method for generating a cardiac activation map. The method comprises ECG data acquisition which can be 12-lead ECG recording and data collection and further a vector-based ECG analysis which comprises evaluation of QRS morphology and timing intervals to estimate arrhythmia origin. The method further comprises preliminary zone identification which comprises determining the likely arrhythmia location using ECG vector components. The method further comprises an optimized regularization process using vector-based constraints to refine the inverse problem solution. The method further comprises solving the inverse and forward ECG problem and thus generating a more accurate and stable electrical activation map. The method further comprises activation map refinement, further improving the map precision with additional constraints. The method further comprises supporting arrhythmia diagnosis and localization for treatment planning by a clinical utility.

[0244] Figure 6 shows a flowchart of a method for noninvasive cardiac mapping. The method comprises ECG acquisition which can be 12-lead ECG data collection and ECG data preprocessing comprising noise reduction and baseline correction. The method further comprises signal analysis and feature extraction and electrophysiological modeling and activation map computation. The method further comprises noninvasive ECG-based cardiac mapping and outputting electroanatom ical activation maps. The method further comprises arrhythmia source identification and treatment planning by a clinical utility.

[0245] Figure 7 shows a flowchart of a method for segmenting cardiac anatomical structures from medical imaging data for use in non-invasive cardiac mapping. The method comprises importing CT I MRI data into a dedicated software and automatic segmentation of cardiac structures with deep learning based on the imported data. The deep learning model is chosen from a model database and is continuously re-trained and updated. The method further comprises non-invasive ECG mapping on the personalized anatomical model based on the segmented cardiac structures.

[0246] The method further comprises merging invasive data to the personalized anatomical model providing enhanced view for navigation and treatment on the personalized anatomical model.

[0247] Figure 8 shows a flowchart of a method for analyzing ECG signals for use in non-invasive cardiac mapping. The method comprises importing ECG data and preprocessing the ECG data by removing baseline wander and noise. The method further comprises arrhythmia classification which can be done with an artificial intelligence algorithm or model. The method further comprises detecting peaks, deflections and separate beats. The step of detection can be done with an artificial intelligence algorithm or model in one step. Figure 9 shows a flowchart of a method for generating a three-dimensional activation map of cardiac tissue based on a 12-lead electrocardiogram (ECG). The method comprises the step of receiving 12-lead ECG data 30 of a patient 1 , the step of selecting at least one cardiac cycle 68 from the ECG data 30 corresponding to an arrhythmic morphology, the step of estimating activation times across cardiac tissue based on the selected cardiac cycle 68, and the step of generating a correlation metric between the selected ECG data 30 and precomputed simulated 12-lead ECG signals stored in a database 80, wherein each simulated 12-lead ECG signal is associated with a simulated activation map 67. The method further comprises selecting at least one simulated 12-lead ECG signal based on the correlation metric and retrieving the associated activation map(s) 67, generating an initial activation map 63 by averaging the activation maps 67 of the selected ECG signals, projecting the initial activation map 63 onto a patient-specific mesh by applying anatomical coordinate mapping between the average heart model and a patient-specific anatomical heart model, the step of generating a final activation map 65 that best fits the measured 12-lead ECG data by optimizing one or more parameters of the initial activation map 63, the one or more parameters including at least one of conduction velocity, fiber orientation, and cardiac anisotropy, and the step of displaying the final activation map 65 using a 3D anatomical representation of the heart.

[0248] Figure 10 shows a flowchart of a further method for generating a three-dimensional activation map of cardiac tissue based on a 12-lead electrocardiogram (ECG). Input data from a patient 1 is acquired using an ECG machine 10, resulting in ECG data 30. From the ECG data 30, a user 2 selects an ECG fragment 66 comprising at least one cardiac cycle 68 that corresponds to an arrhythmic morphology. The selection of the ECG fragment 66 can also be performed automatically without any interaction with the user 2. The selected ECG fragment 66 is compared against a database 80 containing precomputed simulated 12-lead ECG signals, each associated with a simulated activation map 67.

[0249] A loop iterates over the simulated ECG data in the database 80. For each simulated ECG signal, a correlation metric is calculated to determine its similarity to the selected ECG fragment 66. The metric function may include various similarity measures such as correlation coefficient, cross-correlation, Lpnorm, or cosine distance. After computing the similarity scores, a subset of simulated ECG signals with the highest scores is selected. The corresponding simulated activation maps 67 for these signals are retrieved from the database 80. The current implementation selects M=3 maps.

[0250] These selected activation maps 67 are averaged to generate an initial activation map 63. This initial activation map 63 is further optimized to match the measured ECG data 30 by adjusting one or more parameters, such as conduction velocity, fiber orientation, or cardiac anisotropy.

[0251] The result is a final activation map 65. The final activation map 65 is displayed using a three-dimensional anatomical representation of the heart 16.

[0252] Figure 11 shows a flowchart of a further method for generating a three-dimensional activation map of cardiac tissue based on a 12-lead electrocardiogram (ECG). The method comprises the step of receiving 12-lead ECG data 30 of a patient 1 , the step of selecting at least one cardiac cycle 68 from the ECG data 30 corresponding to an arrhythmic morphology, the step of estimating activation times across cardiac tissue based on the selected cardiac cycle 68, and the step of generating a correlation metric between the selected ECG data 30 and precomputed simulated 12-lead ECG signals stored in a database 80, wherein each simulated 12-lead ECG signal is associated with a simulated activation map 67. The method further comprises selecting at least one simulated 12-lead ECG signal based on the correlation metric and retrieving the associated activation map(s) 67, generating an initial activation map 63 by averaging the activation maps 67 of the selected ECG signals, projecting the initial activation map 63 onto a patient-specific mesh by applying anatomical coordinate mapping between the average heart model and a patient-specific anatomical heart model, the step of generating a final activation map 65 that best fits the measured 12-lead ECG data by optimizing one or more parameters of the initial activation map 63, the one or more parameters including at least one of conduction velocity, fiber orientation, and cardiac anisotropy, and the step of displaying the final activation map 65 using a 3D anatomical representation of the heart.

[0253] Figure 10 shows a flowchart of a further method for generating a three-dimensional activation map of cardiac tissue based on a 12-lead electrocardiogram (ECG). Input data from a patient 1 is acquired using an ECG machine 10, resulting in ECG data 30. From the ECG data 30, a user 2 selects an ECG fragment 66 comprising at least one cardiac cycle 68 that corresponds to an arrhythmic morphology. The selection of the ECG fragment 66 can also be performed automatically without any interaction with the user 2. The selected ECG fragment 66 is compared against a database 80 containing precomputed simulated 12-lead ECG signals, each associated with a simulated activation map 67.

[0254] A loop iterates over the simulated ECG data in the database 80. For each simulated ECG signal, a correlation metric is calculated to determine its similarity to the selected ECG fragment 66. The metric function may include various similarity measures such as correlation coefficient, cross-correlation, Lpnorm, or cosine distance.

[0255] After computing the similarity scores, a subset of simulated ECG signals with the highest scores is selected. The corresponding simulated activation maps 67 for these signals are retrieved from the database 80. The current implementation selects M=3 maps.

[0256] These selected activation maps 67 are averaged to generate an initial activation map 63.

[0257] Imaging data 12 including CT and / or MRI is segmented and a mesh of the heart 16 is created to generat a patient-specific parameterized anatomical heart model. The initial activation map 63 is projected onto the patient-specific mesh by applying anatomical coordinate mapping between an average heart model and a patient-specific anatomical heart model.

[0258] This initial activation map 63 is further optimized to match the measured ECG data 30 by adjusting one or more parameters, such as conduction velocity, fiber orientation, or cardiac anisotropy.

[0259] The result is a final activation map 65. The final activation map 65 is displayed using a three-dimensional anatomical representation of the heart 16.

[0260] The first itemized list refers to the aspect relating to a computer-implemented method for time course comparison of 3D electroanatom ic maps for arrhythmia monitoring. The items of the first itemized list can be combined with one or more items of all other itemized lists in this document as well as with one or more features of the claims.

[0261] First itemized list:

[0262] 1. A computer-implemented method for time course comparison of 3D electroanatom ic maps for arrhythmia monitoring, the method comprising: receiving a set of 12-lead ECG data comprising ECG data obtained at at least two time points from a patient; processing the ECG data to generate a 3D electroanatomic map for each of the at least two time points, the 3D electroanatomic map representing cardiac activation; computing quantitative metrics that describe activation characteristics of the heart of the patient for each of the 3D electroanatomic maps; comparing the computed metrics across the at least two time points to identify spatial and / or temporal changes in arrhythmogenic substrates; detecting whether an arrhythmia at a later time point corresponds to a previously treated substrate or a newly emerged substrate; generating a report comprising indicators of treatment efficacy and recurrence risk; and providing said report to a user interface for supporting a physician’s treatment decision.

[0263] 2. A computer-implemented method for time course comparison of electroanatomic maps, the method comprising: receiving a first set of ECG data obtained at a first time point from a patient and a second set of ECG data obtained at a second time point from the patient; generating from the first set of ECG data a first electroanatomic map; generating from the second set of ECG data a second electroanatomic map; determining first quantitative metrics that describe activation characteristics of the heart of the patient for the first electroanatomic map; determining second quantitative metrics that describe activation characteristics of the heart of the patient for the second electroanatomic map; and comparing the first quantitative metrics and the second quantitative metrics.

[0264] 3. The computer-implemented method of item 2, wherein the ECG data is 12-lead ECG data. 4. The computer-implemented method of item 2 or 3, wherein the first electroanatom ic map is a 3D electroanatom ic map and the second electroanatom ic map is a 3D electroanatom ic map.

[0265] 5. The computer-implemented method of one of items 2-4, further comprising detecting whether an arrhythmia at the second time point corresponds to a previously treated substrate or a new arrhythmogenic substrate.

[0266] 6. The computer-implemented method of one of items 2-5, further comprising generating a report considering the comparison of the first quantitative metrics and the second quantitative metrics.

[0267] 7. The computer-implemented method of one of item 6, further comprising providing the report to a user.

[0268] 8. The computer-implemented method of one of items 2-7, wherein the first electroanatom ic map and the second electroanatomic map are generated using inverse solution modeling based on the ECG data.

[0269] 9. The computer-implemented method of one of items 2-8, further comprising applying statistical or machine learning models to classify arrhythmia recurrence patterns.

[0270] 10. The computer-implemented method of one of items 2-9, wherein the quantitative metrics comprise one or more of: activation time, voltage amplitude, conduction velocity, or dominant frequency.

[0271] 11. The computer-implemented method of one of items 2-10, further comprising storing the generated electroanatomic maps and the determined metrics in a patient database for follow-up analysis and / or treatment decision.

[0272] 12. A system for time course comparison of electroanatomic maps, comprising: an input interface configured to receive a first set of ECG data obtained at a first time point from a patient and a second set of ECG data obtained at a second time point from the patient; a processor configured to: generate from the first set of ECG data a first electroanatomic map; generate from the second set of ECG data a second electroanatomic map; determine first quantitative metrics that describe activation characteristics of the heart of the patient for the first electroanatomic map; determine second quantitative metrics that describe activation characteristics of the heart of the patient for the second electroanatomic map; and compare the first quantitative metrics and the second quantitative metrics; and an output interface for outputting a comparison report to a user.

[0273] 13. A computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of any one of items 1- 11.

[0274] 14. A computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the method of any one of items 1-11.

[0275] 15. A data processing apparatus, comprising means for carrying out the method of items 1-11.

[0276] The first itemized list relates to a computer-implemented method and system for automated time course comparison of 3D electroanatom ic maps for the purpose of optimized arrhythmia treatment and follow-up care. The method enables the generation, analysis, and comparison of electroanatom ic maps derived from 12-lead ECG data obtained from a patient at different clinical timepoints. These timepoints may include stages before, during, and after catheter ablation procedures as well as during subsequent follow-up visits. The invention enables improving long-term monitoring of arrhythmogenic activity, providing physicians with a tool that supports consistent and comprehensive management of arrhythmia cases over the entire care cycle.

[0277] An aspect of the disclosed method is the generation of sequential electroanatom ic maps across these timepoints using non-invasive 12-lead ECG input. This allows for consistent, repeatable, and objective monitoring of arrhythmic changes without requiring invasive procedures or complex imaging technologies, while maintaining the accuracy and clinical utility of gold-standard electroanatomic mapping. From the generated maps, quantitative metrics are extracted that characterize cardiac activation, such as activation time, voltage amplitude, conduction velocity, or dominant frequency. These metrics form the basis for a direct comparison of cardiac activation patterns across different timepoints. The method provides the capacity to support physicians in distinguishing whether an arrhythmia detected during a follow-up period is linked to a previously treated substrate or whether it arises from a newly developed pathological region. This differentiation is achieved by comparing the quantified activation characteristics of the 3D maps across timepoints. The system thereby addresses the clinical challenge of monitoring arrhythmia progression or recurrence over time and helps determine whether re-treatment or a new intervention strategy is necessary.

[0278] The invention further enhances clinical decision-making by providing a structured and quantitative basis for evaluating treatment efficacy. Through the automated analysis and comparison of activation metrics, the system supports the adjustment of ablation strategies or medication regimens to directly address the identified sources of arrhythmia. A report is generated that integrates these insights, presenting indicators of treatment success, recurrence risk, and progression trends. This report is then delivered through a user interface to guide the physician’s clinical decisions.

[0279] The system enables comprehensive long-term follow-up. By ensuring consistent comparisons at different timepoints, it allows clinicians to track changes in the electrophysiological state of the heart with high accuracy and objectivity. In doing so, the method supports proactive care strategies that can detect emerging risks early, refine ongoing treatment plans, and ultimately improve patient outcomes.

[0280] The method is cost-effective, leveraging standard 12-lead ECG data instead of relying on resource-intensive imaging or mapping technologies. This makes the system and method more accessible and practical in a variety of clinical settings. Furthermore, the inclusion of features such as the application of statistical or machine learning models for the classification of recurrence patterns, and the storage of map data and associated metrics in a patient database, adds further clinical utility. These elements allow for populationlevel analysis and refinement of treatment protocols over time, supporting evidencebased care decisions.

[0281] The items of the first itemized list address several unmet needs in current arrhythmia management: it enables automated and non-invasive longitudinal monitoring, supports precise differentiation between new and recurring arrhythmias, improves treatment personalization, enhances diagnostic accuracy, facilitates comprehensive follow-up, and promotes cost-effective patient care. By bridging the gap between real-time diagnostics and long-term patient management, the disclosed method provides a robust and scalable solution for clinicians engaged in the treatment and monitoring of arrhythmias.

[0282] The second itemized list refers to the aspect relating to a computer-implemented method for generating a cardiac activation map. The items of the second itemized list can be combined with one or more items of all other itemized lists in this document as well as with one or more features of the claims.

[0283] Second itemized list:

[0284] 1. A computer-implemented method for generating a cardiac activation map, the method comprising: receiving 12-lead ECG data of a patient; performing vector-based analysis on the ECG data to identify an approximate region of origin of an arrhythmia, wherein the analysis is based on QRS morphology and timing intervals; generating the cardiac activation map by solving an inverse ECG problem, wherein solving the problem comprises a regularization procedure and wherein the identified approximate region of origin is used as a constraint in the regularization procedure; and outputting the cardiac activation map for clinical interpretation.

[0285] 2. A computer-implemented method for generating a cardiac activation map, the method comprising: receiving ECG data of a patient; identifying an approximate region of origin of an arrhythmia; and generating the cardiac activation map based on the identified approximate region.

[0286] 3. The computer-implemented method of item 2, wherein ECG data is 12-lead ECG data.

[0287] 4. The computer-implemented method of item 3, wherein the 12-lead ECG data is obtained during an arrhythmic episode. 5. The computer-implemented method of one of items 2-4, wherein identifying the approximate region of origin comprises performing a vector-based analysis on the ECG data.

[0288] 6. The computer-implemented method of item 5, wherein the vector-based analysis is based on QRS morphology and / or timing intervals.

[0289] 7. The computer-implemented method of one of items 2-6, further comprising outputting the cardiac activation map.

[0290] 8. A system for generating a cardiac activation map, the system comprising: an input interface configured to receive ECG data; a processor configured to: identify an approximate region of origin of an arrhythmia; and generate the cardiac activation map based on the identified approximate region; and an output interface for outputting the activation map to a user.

[0291] 9. A computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of any one of items 1- 7.

[0292] 10. A computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the method of any one of items 1-7.

[0293] 11. A data processing apparatus, comprising means for carrying out the method of items 1-7.

[0294] The items of the second itemized list relate to a method for generating robust electrical activation maps of the heart by solving inverse electrocardiographic (ECG) problems using vector-based regularization derived from 12-lead ECG data. The method relates to a process that utilizes vector-based ECG information to optimize the computational resolution of inverse problems in the context of cardiac mapping.

[0295] Traditionally, the inverse ECG problem — which involves calculating an internal cardiac activation map based on noninvasive surface measurements such as those from a 12- lead ECG — has been characterized as mathematically ill-posed. This means that small measurement errors can lead to large inaccuracies in the calculated map, particularly when physiological or anatomical constraints are not available or are insufficiently defined. As a result, regularization methods used in the solution of the inverse problem often fail to yield reliable or clinically meaningful results.

[0296] The items introduce a novel approach to overcome these limitations by incorporating a vector-based analysis step prior to the solution of the inverse problem. The method applies well-established principles of ECG vector theory to the 12-lead ECG data to determine the approximate region of origin of an arrhythmia. This analysis focuses in particular on interpreting QRS morphology and timing intervals, allowing the system to infer the spatial zone of the arrhythmic source. The resulting information is then used as a constraint within the regularization step of the inverse problem.

[0297] By integrating this vector-based localization into the computational pipeline, the inverse problem is transformed from an under-determined system into one that is more tractable and bounded by clinically relevant parameters. The constrained regularization procedure thereby benefits from improved robustness and convergence properties, leading to enhanced accuracy and precision in the generation of activation maps.

[0298] The technical implementation comprises three aspects: (1) the integration of 12-lead ECG vector theory into the analysis process to derive a physiological estimation of arrhythmia origin, (2) the use of this estimation to guide and constrain the regularization procedure when solving the inverse problem, and (3) the generation of cardiac activation maps with improved spatial resolution and clinical reliability.

[0299] This method addresses the inherent instability of ill-posed inverse ECG problems by narrowing the solution space using physiologically meaningful constraints. It offers a noninvasive, cost-effective, and efficient alternative to catheter-based invasive mapping procedures, thereby reducing patient risk and resource burden. It supports clinical workflows by enabling faster identification of arrhythmic zones, which streamlines diagnosis and therapeutic planning. Furthermore, it introduces a data-driven enhancement to standard regularization techniques, moving beyond generic mathematical assumptions toward patient-specific, physiology-informed constraints.

[0300] The method exhibits broad applicability across different types of arrhythmias and patient anatomies, making it a versatile and scalable tool for diverse clinical environments. By enabling the creation of more reliable and accurate electrical activation maps from standard surface ECG data, the method improves the clinical utility of noninvasive cardiac mapping. The third itemized list refers to the aspect relating to a computer-implemented method for noninvasive cardiac mapping. The items of the third itemized list can be combined with one or more items of all other itemized lists in this document as well as with one or more features of the claims.

[0301] Third itemized list:

[0302] 1. A computer-implemented method for noninvasive cardiac mapping, the method comprising: receiving 12-lead ECG data of a patient; preprocessing the ECG data comprising noise reduction and baseline correction; selecting an average heart model from a set of pre-defined anatomical models based on parameters of the patient including gender and estimated heart position within the torso; generating an electroanatom ical activation map of the heart by processing the ECG data using the selected heart model and an electrophysiological simulation; identifying a region of origin and propagation pathways of the arrhythmia based on the generated activation map; and outputting the electroanatomical activation map, the identified region of origine and the identified propagation pathways for clinical interpretation.

[0303] 2. A computer-implemented method for generating a cardiac activation map, the method comprising: receiving ECG data of a patient; generating an electroanatomical activation map of the heart by processing the ECG data using a heart model and an electrophysiological simulation; and identifying a region of origin and / or propagation pathways of the arrhythmia based on the generated activation map.

[0304] 3. The computer-implemented method of item 2, wherein the ECG data is 12-lead ECG data.

[0305] 4. The computer-implemented method of item 3, wherein the 12-lead ECG data is obtained during an arrhythmic episode.

[0306] 5. The computer-implemented method of item 4, wherein only 12-lead ECG data is used as input data for generating the electroanatomical activation map. 6. The computer-implemented method of items 2-5, further comprising preprocessing the ECG data.

[0307] 7. The computer-implemented method of item 6, wherein the preprocessing comprises noise reduction and / or baseline correction.

[0308] 8. The computer-implemented method of items 2-7, further comprising selecting an average heart model from a set of pre-defined anatomical models, wherein the selected model is used for generating the electroanatomical activation map.

[0309] 9. The computer-implemented method of item 8, wherein selecting the heart model is based on parameters of the patient including gender and / or estimated heart position within the torso.

[0310] 10. The computer-implemented method of items 8 or 9, wherein the heart model is selected from a database containing gender-specific anatomical variations.

[0311] 11. A system for noninvasive cardiac mapping, comprising: an input interface configured to receive ECG data; a processor configured to: generate an electroanatomical activation map of the heart by processing the ECG data using a heart model and an electrophysiological simulation; and identify a region of origin and / or propagation pathways of the arrhythmia based on the generated activation map; and an output interface for outputting the activation map to a user.

[0312] 12. A computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of any one of items 1- 10.

[0313] 13. A computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the method of any one of items 1-10.

[0314] 14. A data processing apparatus, comprising means for carrying out the method of items 1-10.

[0315] The items of the third itemized list concern a method for noninvasive cardiac mapping that computes electroanatomical activation maps of cardiac arrhythmias using only data derived from standard 12-lead electrocardiograms (ECG), without requiring imaging modalities such as computed tomography (CT) or magnetic resonance imaging (MRI). This method is implemented through a novel algorithm that integrates electrophysiological modeling with advanced signal processing and computational techniques. It is designed to provide physicians with a rapid, accurate, and cost-effective diagnostic tool for identifying the origins and propagation pathways of cardiac arrhythmias, including both focal and reentrant types.

[0316] A feature of the method is its tomography-free approach. Unlike conventional mapping techniques that require imaging data to reconstruct individualized anatomical models, the invention employs an average heart model as the basis for its computations. This strategy eliminates or reduces the dependence on costly and resource-intensive imaging systems, making the method widely deployable, particularly in settings where advanced imaging tools are not available. The algorithm further enhances data interpretation by incorporating individualized parameters such as the patient’s gender and the anatomical position of the heart within the torso, which contributes to the accuracy and personalization of the mapping output.

[0317] The method yields electroanatomical activation maps that visually represent the spatial and temporal characteristics of electrical signal propagation in the heart. These outputs are specifically designed to assist clinicians in diagnosing arrhythmia sources and understanding their behavior, thereby facilitating informed and timely therapeutic decisions. By presenting the results in a user-friendly format, the algorithm supports swift clinical evaluation without requiring additional interpretation infrastructure.

[0318] The method addresses several challenges in the field of cardiac electrophysiology. It enhances accessibility by relying exclusively on ECG systems, which are widely available even in resource-limited healthcare environments. It reduces the cost associated with arrhythmia diagnostics and treatment by removing the need for expensive imaging procedures. It delivers results rapidly, promoting efficient clinical workflows. The method achieves a high level of precision through the integration of gender-specific physiological characteristics and spatial modeling of the heart's position, leading to reliable and reproducible diagnostic outcomes. The method contributes to improved clinical outcomes by equipping physicians with detailed visualizations of arrhythmia mechanisms, thus supporting more targeted and effective treatment strategies. The fourth itemized list refers to the aspect relating to a computer-implemented method for segmenting cardiac anatomical structures from medical imaging data for use in non- invasive cardiac mapping. The items of the fourth itemized list can be combined with one or more items of all other itemized lists in this document as well as with one or more features of the claims.

[0319] Fourth itemized list:

[0320] 1. A computer-implemented method for segmenting cardiac anatomical structures from medical imaging data for use in non-invasive cardiac mapping, the method comprising: receiving a cardiac medical image of a patient; identifying and labeling patient-specific cardiac structures in the medical image using a trained deep learning model, wherein the cardiac structures include at least one of the left and right ventricular walls, blood pools, and right ventricular outflow tract; generating an annotated volumetric cardiac image including segmentation labels for the identified cardiac structures; modifying the segmentation labels using a graphical user interface based on an input of a user; resampling the annotated volumetric image to a desired spatial resolution; and generating a volumetric or surface mesh from the resampled image for use in computational electrophysiology.

[0321] 2. A computer-implemented method for segmenting cardiac anatomical structures from medical imaging data, the method comprising: receiving a cardiac medical image of a patient; identifying patient-specific cardiac structures in the medical image using a trained deep learning model; and generating an annotated volumetric cardiac image based on the identified cardiac structures.

[0322] 3. The computer-implemented method of item 2, wherein the cardiac medical image is obtained using at least one of MR, CT, or ultrasound.

[0323] 4. The computer-implemented method of items 2 or 3, wherein identifying the patient-specific cardiac structures comprises labeling the patient-specific cardiac structures. 5. The computer-implemented method of one of items 2-4, wherein the cardiac structures include at least one of the left and right ventricular walls, blood pools, and right ventricular outflow tract.

[0324] 6. The computer-implemented method of one of items 2-5, wherein the annotated volumetric cardiac image includes segmentation labels for the identified cardiac structures.

[0325] 7. The computer-implemented method of item 6, further comprising modifying the segmentation labels.

[0326] 8. The computer-implemented method of item 7, wherein modifying the segmentation labels is based on an input of a user.

[0327] 9. The computer-implemented method of one of items 2-8, further comprising resampling the annotated volumetric image to a desired spatial resolution.

[0328] 10. The computer-implemented method of one of items 2-9, further comprising generating a volumetric or surface mesh from the annotated volumetric cardiac image for use in computational electrophysiology.

[0329] 11. The computer-implemented method of one of items 2-10, wherein the deep learning model is retrained over time using an updated dataset of medical images with user-verified annotations.

[0330] 12. The computer-implemented method of one of items 2-11 , wherein a graphical interface allows the user to add, delete, or relabel anatomical segmentations.

[0331] 13. A system for segmenting cardiac anatomical structures from medical imaging data, comprising: an input interface configured to receive cardiac medical imaging data; a processor configured to: identify the patient-specific cardiac structures in the medical image using a trained deep learning model; and generate an annotated volumetric cardiac image based on the identified cardiac structures; and an output interface for outputting the annotated volumetric cardiac image.

[0332] 14. A computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of any one of items 1- 12. 15. A computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the method of any one of items 1-12.

[0333] 16. A data processing apparatus, comprising means for carrying out the method of items 1-12.

[0334] The items of the fourth itemized list relate to a system and method for cardiac segmentation tailored to applications in non-invasive cardiac mapping, particularly electrocardiographic (ECG) imaging. ECG imaging estimates the electrical activity of the heart based on electrical potentials measured on the surface of the torso of the patient. While torso anatomy can be captured non-invasively, e.g., via photographic acquisition, the accurate anatomical modeling of the heart remains a challenge as it necessitates detailed medical imaging (MRI, CT, ultrasound, etc.). Current cardiac segmentation methods are semi-automatic, require expert intervention, and typically deliver incomplete anatomical models, such as focusing exclusively on the left ventricular wall.

[0335] To address these limitations, the disclosed method and system introduce a fully automatic pipeline for cardiac segmentation, based on a deep learning model trained to process medical image data and output volumetric images with labeled cardiac anatomical structures. The system identifies multiple distinct cardiac components, including the left and right ventricular walls, the right ventricular outflow tract, and blood pools. The model is trained on progressively expanding annotated datasets, where new cases continuously improve the robustness and accuracy of the segmentation.

[0336] The system further includes functionality allowing physicians to review, modify, or accept the automatically produced annotations. The segmented output can be resampled to conform to the spatial resolution needed for numerical computations in electrophysiology, and converted into surface or volumetric mesh formats suitable for simulation. This enables direct integration with patient-specific ECG imaging applications. Electrical signals measured at the torso surface can be mapped onto the segmented heart and its substructures to produce electroanatomical maps reflecting actual physiological function.

[0337] Additionally, the system and method support merging with data from invasive electrophysiological systems or structural imaging data from modalities such as GE-MRI, allowing comprehensive integration of scar tissue imaging and other clinically relevant metrics into the model. These enriched data inputs enhance both the accuracy of simulations and the value of resulting visualizations for clinical applications.

[0338] This solution meets several needs in the domain of cardiac electrophysiology. First, it addresses the lack of tools for the fully automatic segmentation of all relevant cardiac structures in ECG imaging. Second, it reduces the time and labor required to produce patient-specific anatomical models suitable for electrical field simulation, which historically has been a barrier to the scalability and adoption of ECG imaging. Third, it allows faster clinical decision-making by minimizing the time lag between diagnostic imaging and the delivery of electroanatom ical maps.

[0339] Overall, this solution improves the accuracy, efficiency, and clinical integration of non- invasive cardiac mapping by providing a scalable and automated means of cardiac structure segmentation and model generation.

[0340] The fifth itemized list refers to the aspect relating to a computer-implemented method for analyzing ECG signals for use in non-invasive cardiac mapping. The items of the fifth itemized list can be combined with one or more items of all other itemized lists in this document as well as with one or more features of the claims.

[0341] Fifth itemized list:

[0342] 1. A computer-implemented method for analyzing ECG signals for use in non- invasive cardiac mapping, comprising: receiving 12-lead ECG data from a patient; identifying an underlying cardiac rhythm from the ECG data, the cardiac rhythm including at least one of ventricular tachycardia, sinus rhythm, or atrial fibrillation; selecting by a user of an ECG fragment from the ECG data; classifying the selected ECG fragment as atrial or ventricular; determining time boundaries of one complete event of cardiac activation within the ECG fragment, wherein cardiac activation comprises whole-heart or two-chamber depolarization; and outputting an interval of the ECG data for use in ECG imaging or electroanatom ical map reconstruction, wherein the interval is characterized by the determined time boundaries.

[0343] 2. A computer-implemented method for analyzing ECG signals for use in non- invasive cardiac mapping, comprising: receiving ECG data from a patient; selecting an ECG fragment from the ECG data; and determining time boundaries of one complete event of cardiac activation within the ECG fragment.

[0344] 3. The computer-implemented method of item 2, wherein the ECG data is 12-lead ECG data.

[0345] 4. The computer-implemented method of item 2 or 3, wherein selecting an ECG fragment from the ECG data is based on an input of a user.

[0346] 5. The computer-implemented method of one of items 2-4, further comprising identifying an underlying cardiac rhythm from the ECG data.

[0347] 6. The computer-implemented method of item 5, wherein the cardiac rhythm includes at least one of ventricular tachycardia, sinus rhythm, or atrial fibrillation.

[0348] 7. The computer-implemented method of one of items 2-6, further comprising classifying the selected ECG fragment as atrial or ventricular.

[0349] 8. The computer-implemented method of one of items 2-7, wherein cardiac activation includes one of whole-heart or two-chamber depolarization.

[0350] 9. The computer-implemented method of one of items 2-8, further comprising outputting an interval of the ECG data.

[0351] 10. The computer-implemented method of item 9, wherein the interval is characterized by the determined time boundaries.

[0352] 11. A system for analyzing ECG signals for use in non-invasive cardiac mapping, the system comprising: an input interface configured to receive ECG data; a processor configured to: select an ECG fragment from the ECG data; and determine time boundaries of one complete event of cardiac activation within the ECG fragment; and an output interface for outputting the identified ECG interval. 12. A computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of any one of items 1- 10.

[0353] 13. A computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the method of any one of items 1-10.

[0354] 14.A data processing apparatus, comprising means for carrying out the method of items 1-10.

[0355] The items of the fifth itemized list relate to an automated method for analyzing and delineating electrocardiogram (ECG) signals to enhance the accuracy of non-invasive cardiac mapping or ECG imaging (ECGI). A challenge in ECGI lies in the precise identification of ECG intervals that correspond to full cycles of cardiac electrical activation. Selection of inappropriate intervals — either too brief or too prolonged — can result in incomplete or artificially extended representations of electrical activity, thereby compromising the reliability of reconstructed electroanatomical maps.

[0356] This solution introduces a systematic method for automatically selecting ECG intervals that correspond to physiologically complete cardiac activation events. The method comprises the classification of the underlying cardiac rhythm (e.g., normal sinus rhythm, atrial fibrillation, ventricular tachycardia), based on an analysis of the ECG signal. For a selected ECG fragment, the method automatically determines whether the fragment represents atrial or ventricular activation and identifies the precise boundaries of the interval that encapsulates one full event of electrical depolarization, either across the whole heart or confined to one of its chambers.

[0357] The resulting delineated interval serves as a consistent, beat-to-beat input for ECGI, aligning the cardiac state with reconstructed electroanatomical maps and eliminating distortions that may arise from missing or excessive timestamp data. This enhances the anatomical and physiological fidelity of non-invasive cardiac imaging.

[0358] The solution offers multiple advantages over conventional approaches in cardiac electrophysiology. The solution automates the process of ECG interval selection, mitigating inter-operator variability and yielding reproducible input for downstream ECG analyses. The solution improves the integrity of ECGI outcomes by ensuring that selected intervals conform to the assumed excitation propagation models, thereby enhancing the interpretability of reconstructed maps. The solution streamlines the ECGI workflow by reducing manual intervention, leading to faster processing times and contributing to lower healthcare-related costs.

[0359] By enabling precise, automated, and rhythm-specific delineation of ECG intervals, the invention addresses limitations in current non-invasive cardiac mapping technologies and supports more accurate and efficient assessments of cardiac electrical function.

[0360] The sixth itemized list refers to a method for generating a three-dimensional activation map of cardiac tissue based on a 12-lead electrocardiogram (ECG). The items of the sixth itemized list can be combined with one or more items of all other itemized lists in this document as well as with one or more features of the claims.

[0361] Sixth itemized list:

[0362] 1. A computer-implemented method for generating a three-dimensional activation map of cardiac tissue based on a 12-lead electrocardiogram (ECG), the method comprising: receiving 12-lead ECG data of a patient; selecting at least one cardiac cycle from the ECG data corresponding to an arrhythmic morphology; estimating activation times across cardiac tissue based on the selected cardiac cycle; generating a correlation metric between the selected ECG data and precomputed simulated 12-lead ECG signals a database, wherein each simulated 12-lead ECG signal is associated with a simulated activation map; selecting at least one simulated 12-lead ECG signal based on the correlation metric and retrieving the associated activation map(s); generate an initial activation map by averaging the activation maps of the selected ECG signals; generating a final activation map that best fits the measured 12-lead ECG data by optimizing one or more parameters of the initial activation map, the one or more parameters including at least one of conduction velocity, fiber orientation, and cardiac anisotropy; and displaying the final activation map using a 3D anatomical representation of the heart.

[0363] 2. The computer-implemented method of item 1 , further comprising projecting the initial activation map onto a patient-specific mesh by applying anatomical coordinate mapping between the average heart model and a patient-specific anatomical heart model.

[0364] 3. A computer-implemented method for generating a cardiac activation map, the method comprising: receiving data for at least one cardiac cycle obtained from ECG data of a patient; determining at least one parameter representing cardiac activation based on the received data; obtaining from a database at least one activation map, wherein the obtained at least one activation map is selected from the database based on the determined at least one parameter; generating an initial activation map based on the at least one activation map obtained from the database; and generating a final activation map by optimizing the initial activation map.

[0365] 4. The computer-implemented method of item 3, further comprising receiving ECG data of the patient.

[0366] 5. The computer-implemented method of item 3 or 4, wherein the ECG data is 12- lead ECG data.

[0367] 6. The computer-implemented method of item 4 or 5, further comprising selecting the at least one cardiac cycle from the ECG data.

[0368] 7. The computer-implemented method of item 7, wherein the at least one cardiac cycle is selected corresponding to an arrhythmic morphology.

[0369] 8. The computer-implemented method of one of items 3-7, wherein the at least one parameter representing cardiac activation is activation times across cardiac tissue.

[0370] 9. The computer-implemented method of one of items 3-8, wherein obtaining the at least one activation map from the database comprises comparing the selected at least one cardiac cycle to ECG data stored in the database.

[0371] 10. The computer-implemented method of item 9, wherein the ECG data stored in the database comprises precomputed simulated ECG signals. 11. The computer-implemented method of one of items 9 and 10, wherein the step of comparing comprises generating a correlation metric between the selected at least one cardiac cycle and at least one of the precomputed simulated ECG signals.

[0372] 12. The computer-implemented method of item 11 , wherein the step of comparing further comprises selecting at least one precomputed simulated ECG signal based on the correlation metric and obtain the at least one activation map which is associated with the selected at least one precomputed simulated ECG signal.

[0373] 13. The computer-implemented method of one of items 3-12, wherein generating the initial activation map comprises averaging the at least one obtained activation maps.

[0374] 14. The computer-implemented method of one of items 3-13, wherein generating the final activation map comprises optimizing one or more parameters of the initial activation map.

[0375] 15. The computer-implemented method of one of item 14, wherein the one or more parameters include at least one of conduction velocity, fiber orientation, and cardiac anisotropy.

[0376] 16. The computer-implemented method of one of items 3-15, further comprising outputting the activation map to a user.

[0377] 17. The computer-implemented method of one of items 3-15, further comprising projecting the initial activation map onto a patient-specific model of the heart.

[0378] 18. A system for generating a cardiac activation map, comprising: an input interface configured to receive data for at least one cardiac cycle obtained from ECG data of a patient; a database configured for storing at least one activation map; a processor configured to: determine at least one parameter representing cardiac activation based on the received data; obtain from the database at least one activation map, wherein the obtained at least one activation map is selected from the database based on the determined at least one parameter; generate an initial activation map based on the at least one activation map obtained from the database; and generate a final activation map by optimizing the initial activation map. 19. The system of item 18, further comprising an output interface for outputting the activation map to a user.

[0379] 20. A computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of any one of items 1- 17.

[0380] 21. A computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the method of any one of items 1-17.

[0381] 22. A data processing apparatus, comprising means for carrying out the method of items 1-17.

[0382] The seventh itemized list refers to method for generating a dataset of simulated cardiac activation patterns. The items of the seventh itemized list can be combined with one or more items of all other itemized lists in this document as well as with one or more features of the claims.

[0383] Seventh itemized list:

[0384] 1. A computer-implemented method for generating a dataset of simulated cardiac activation patterns and corresponding 12-lead ECG signals, the method comprising: providing a three-dimensional anatomical model of the heart and torso, wherein the heart is represented by a tetrahedral mesh incorporating myocardial fiber orientations and tissue conductivities; defining a set of simulation parameters comprising activation source locations and conduction velocity tensors; simulating cardiac activation by solving an anisotropic eikonal equation on the mesh to compute activation times across the myocardium, using a focal excitation at a defined source location; modulating conduction velocities in accordance with the defined tensors to reflect anisotropic propagation, including pathological regions with reduced conduction velocity to mimic scar or fibrotic tissue; repeating simulating cardiac activation for multiple variations of the simulation parameters according to a quasi-random Halton sampling scheme; for each simulation, computing corresponding synthetic 12-lead ECG signals using a lead field method applied to the calculated activation data; storing, for each simulation, the activation map, the associated simulated ECG signals, and metadata describing the simulation parameters. A computer-implemented method for generating a dataset of activation maps and corresponding ECG signals, the method comprising: generating a first cardiac activation map based on an electrical activation model for a first set of simulation parameters; generating a second cardiac activation map based on an electrical activation model for a second set of simulation parameters, wherein the first set is different from the second set; generating first ECG signals corresponding to the first cardiac activation map; and generating second ECG signals corresponding to the second cardiac activation map. The computer-implemented method of item 2, further comprising storing the first cardiac activation map, the second cardiac activation map, the first ECG signals, the second ECG signals, the first set of simulation parameters and the second set of simulation parameters in a database. A system for generating a dataset of activation maps and corresponding ECG signals, comprising: a database configured for storing at the activation maps, the corresponding ECG signals and the corresponding sets of simulation parameters; and a processor configured to: generate a first cardiac activation map based on an electrical activation model for a first set of simulation parameters; generate a second cardiac activation map based on an electrical activation model for a second set of simulation parameters, wherein the first set is different from the second set; generate first ECG signals corresponding to the first cardiac activation map; and generate second ECG signals corresponding to the second cardiac activation map. A computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of any one of items 1- 3. A computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the method of any one of items 1-3. 7. A data processing apparatus, comprising means for carrying out the method of items 1-3.

[0385] DEFINITIONS

[0386] A "patient" designates a human being.

[0387] The verb to "personalize" (in the sense of personalize, personalise, individualize, individualise) can be replaced by: make personal or more personal

[0388] An "ECG machine" designates a device that is capable of receiving standard 12-lead ECG data from a patient and generating a standard 12-lead ECG.

[0389] An "ultrasound machine" designates any device that can be used to generate ultrasound images of the heart.

[0390] A "data integration hub computer" designates a computer that is capable of receiving data from multiple sources, and integrating that data into a single, unified view.

[0391] A "computer screen" designates a display device, such as a computer monitor, a tablet, a smartphone, or a head-mounted display.

[0392] A "3D heart model representation" designates a representation of the heart in 3D space.

[0393] A "3D modeling software" designates software that can be used to create 3D models of the heart,.

[0394] A "segmentation module" designates a module that performs segmentation of cardiac structures from the multi-modality imaging data.

[0395] An "ECG analysis module" designates the algorithms that process the ECG data to calculate the propagation of electrical excitation across the heart. As used herein, the term "deep learning networks" refers to neural networks that use artificial intelligence (Al) techniques to model cardiac structures.

[0396] As used herein, the term "cardiovascular structures" refers to any structure in the heart that may be altered by electrical excitation.

[0397] As used herein, the term "invasive electrophysiological data" refers to any cardiac excitation that causes electrical activity in the heart.

[0398] As used herein, the term "heart models" refers to a collection of cardiac structures that are modelled using imaging data.

[0399] As used herein, the term "advanced algorithms" refers to algorithms that use artificial intelligence (Al) and machine learning techniques (e.g., fuzzy logic), deep learning networks, and other neural networks to process cardiac excitation data in real-time.

[0400] A "personalized 3D model" designates a model of the heart that is personalized to a particular patient.

[0401] As used herein, the term "panoramic electroanatomical maps" refers to a detailed representation of the heart including at least one of activation maps, propagation maps, etc.

[0402] As used herein, the term "deep learning networks for segmentation" refers to a neural network that utilizes machine learning techniques to automatically process cardiac structures from imaging data.

[0403] A "personalized 3D heart model with invasive electrophysiological data" designates a personalized heart model that is augmented with invasive electrophysiological data.

[0404] A "heart model for healthcare professionals" designates a model of the heart that is personalized for a particular patient. 2EPQ 010-IC

[0405] 69

[0406] As used herein, the term "areas of fibrotic tissue" refers to areas of fibrous tissue that may be identified by imaging.

[0407] As used herein, the term "electrical excitation" refers to a propagation of electrical signals across the heart.

[0408] As used herein, the term "existing clinical workflows" refers to existing workflows that have a variety of functions, such as those associated with cardiac diagnostic analysis.

[0409] A "personalized 3D heart model" designates a model of the heart that is personalized to a particular patient, and the term personalized 3D heart model is used interchangeably with the term personalized 3D heart model.

[0410] A "healthcare professional" designates a physician, nurse, physician assistant, or other healthcare professional.

[0411] As used herein, the term "fibrotic tissue" refers to cardiac structures that have been damaged or deteriorated.

[0412] As used herein, the term "cardiovascular electrical sources" refers to any source of electrical energy that can be applied to a heart muscle.

[0413] A "cardiac diagnostic analysis" designates the analysis of the heart using the personalized 3D models.

[0414] A "standard 12-lead ECG datum" designates a set of 12-lead ECG data that has been recorded in a standard manner.

[0415] A "calculated excitation" designates the propagation of electrical excitation across the heart, which is calculated by the advanced algorithms.

[0416] An "activation map" designates an activation map with local timings of the electrical excitation across the heart. A "propagation map" designates a map of the propagation of electrical excitation across the heart.

[0417] A "phase map" designates a map of the phase of the electrical excitation wavefronts in the heart.

[0418] A "rotor density map" designates a map of the number of rotors per unit volume of the heart.

[0419] A "cardiac procedure" designates any procedure that involves the heart, including but not limited to cardiac catheterization, cardiac ablation, cardiac surgery, and cardiac pacing.

[0420] An "impedance measurement" designates the measurement of the electrical impedance of the heart.

[0421] A "catheter position" designates the location of the catheter tip within the heart.

[0422] A "created mesh" designates a mesh is created from the voxel model (result of segmentation) and is used for boundary element solver as an input.

[0423] As used herein, the term "web-based platform" refers to a platform that is hosted on a server or computer and provides access to data in real-time.

[0424] As used herein, the term "finite element solver" refers to a mathematical solution that uses two or more elements as boundary points.

[0425] A "body surface" designates the surface of the body that is in contact with the skin. Protection may be sought for combinations of features which are disclosed in the referenced earlier patent applications DE102024114031.3 of 20 May 2024 and EP25177220.8 of 19 May 2025, the contents of which are herein incorporated by reference. It is disclosed there how these features combinations contribute to achieving the technical aim of the present application and they are thus comprised in the solution of the technical problem underlying the subject matter of the present application. The features and combinations which are disclosed in the reference documents implicitly belong to the description of the subject matter in the present application and thus to the content of the present application as filed.

[0426] REFERENCE NUMERAL LIST

[0427] 1 patient

[0428] 2 user

[0429] 10 ECG machine

[0430] 11 MRI

[0431] 12 CT machine

[0432] 12 CT

[0433] 12 imaging data including CT

[0434] 13 ultrasound machine

[0435] 13 ultrasound

[0436] 14 data integration hub computer

[0437] 15 computer screen

[0438] 16 3D heart model representation

[0439] 16 heart

[0440] 17 3D modeling software

[0441] 20 segmentation module

[0442] 21 ECG analysis module

[0443] 25 maps module

[0444] 30 ECG data

[0445] 32 set of ECG data

[0446] 34 first set of ECG data

[0447] 36 second set of ECG data

[0448] 40 first electroanatom ic map

[0449] 42 second electroanatomic map

[0450] 44 first quantitative metrics

[0451] 46 second quantitative metrics

[0452] 50 report

[0453] 60 approximate region of origin

[0454] 62 cardiac activation map

[0455] 63 initial activation map

[0456] 64 cardiac image

[0457] 65 final activation map

[0458] 66 ECG fragment

[0459] 67 simulated activation map 68 cardiac cycle

[0460] 70 input interface

[0461] 72 processor

[0462] 74 output interface 80 database

Claims

CLAIMS1. A computer-implemented method for generating a three-dimensional activation map of cardiac tissue based on a 12-lead electrocardiogram (ECG), the method comprising: receiving 12-lead ECG data of a patient; selecting at least one cardiac cycle from the ECG data corresponding to an arrhythmic morphology; estimating activation times across cardiac tissue based on the selected cardiac cycle; generating a correlation metric between the selected ECG data and precomputed simulated 12-lead ECG signals stored in a database, wherein each simulated 12-lead ECG signal is associated with a simulated activation map; selecting at least one simulated 12-lead ECG signal based on the correlation metric and retrieving the associated activation map(s); generate an initial activation map by averaging the activation maps of the selected ECG signals; generating a final activation map that best fits the measured 12-lead ECG data by optimizing one or more parameters of the initial activation map, the one or more parameters including at least one of conduction velocity, fiber orientation, and cardiac anisotropy; and displaying the final activation map using a 3D anatomical representation of the heart.

2. The computer-implemented method of claim 1 , further comprising projecting the initial activation map onto a patient-specific mesh by applying anatomical coordinate mapping between the average heart model and a patient-specific anatomical heart model.

3. A computer-implemented method for generating a cardiac activation map, the method comprising: receiving data for at least one cardiac cycle obtained from ECG data of a patient; determining at least one parameter representing cardiac activation based on the received data; obtaining from a database at least one activation map, wherein the obtained at least one activation map is selected from the database based on the determined at least one parameter; generating an initial activation map based on the at least one activation map obtained from the database; and generating a final activation map by optimizing the initial activation map.

4. The computer-implemented method of claim 3, further comprising receiving ECG data of the patient.

5. The computer-implemented method of claim 3 or 4, wherein the ECG data is 12- lead ECG data.

6. The computer-implemented method of claim 4 or 5, further comprising selecting the at least one cardiac cycle from the ECG data.

7. The computer-implemented method of claim 7, wherein the at least one cardiac cycle is selected corresponding to an arrhythmic morphology.

8. The computer-implemented method of one of claims 3-7, wherein the at least one parameter representing cardiac activation is activation times across cardiac tissue.

9. The computer-implemented method of one of claims 3-8, wherein obtaining the at least one activation map from the database comprises comparing the selected at least one cardiac cycle to ECG data stored in the database.

10. The computer-implemented method of claim 9, wherein the ECG data stored in the database comprises precomputed simulated ECG signals.

11. The computer-implemented method of one of claims 9 and 10, wherein the step of comparing comprises generating a correlation metric between the selected at least one cardiac cycle and at least one of the precomputed simulated ECG signals.

12. The computer-implemented method of claim 11 , wherein the step of comparing further comprises selecting at least one precomputed simulated ECG signal based on the correlation metric and obtain the at least one activation map which is associated with the selected at least one precomputed simulated ECG signal.

13. The computer-implemented method of one of claims 3-12, wherein generating the initial activation map comprises averaging the at least one obtained activation maps.

14. The computer-implemented method of one of claims 3-13, wherein generating the final activation map comprises optimizing one or more parameters of the initial activation map.

15. The computer-implemented method of one of claims 14, wherein the one or more parameters include at least one of conduction velocity, fiber orientation, and cardiac anisotropy.

16. The computer-implemented method of one of claims 3-15, further comprising outputting the activation map to a user.

17. The computer-implemented method of one of claims 3-15, further comprising projecting the initial activation map onto a patient-specific model of the heart.

18. A system for generating a cardiac activation map, comprising: an input interface configured to receive data for at least one cardiac cycle obtained from ECG data of a patient; a database configured for storing at least one activation map; a processor configured to: determine at least one parameter representing cardiac activation based on the received data; obtain from the database at least one activation map, wherein the obtained at least one activation map is selected from the database based on the determined at least one parameter; generate an initial activation map based on the at least one activation map obtained from the database; and generate a final activation map by optimizing the initial activation map.

19. The system of claim 18, further comprising an output interface for outputting the activation map to a user.

20. A computer program comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method of any one of claims 1-17.

21. A computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the method of any one of claims 1-17.

22. A data processing apparatus, comprising means for carrying out the method of claims 1-17.

Images