Karli Gillette
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Karli Gillette has a total of 95 co-authors. Here are the top co-authors by number of shared publications:
- Gernot Plank
Gottfried Schatz Research Center: Division of Medical Physics and Biophysics, Medical University of Graz, Graz, Austria
20 shared publications - Anton J. Prassl
Gottfried Schatz Research Center, Division of Medical Physics and Biophysics, Medical University of Graz, Graz, Austria
6 shared publications - Matthias A. F. Gsell
Gottfried Schatz Research Center, Division of Medical Physics and Biophysics, Medical University of Graz, Graz, Austria
6 shared publications - Christoph M. Augustin
Gottfried Schatz Research Center, Division of Biophysics Medical University of Graz Graz Austria
5 shared publications - Edward Joseph Vigmond
Univ. Bordeaux, CNRS, Bordeaux INP, IMB, UMR 5251, IHU Liryc, F-33400 Talence, France
5 shared publications - Jake A. Bergquist
Scientific Computing and Imaging Institute, University of Utah, SLC, UT, USA
5 shared publications
Latest Publications
- Open Access
Explainable Deep Learning-based Classification of Wolff-Parkinson-White Electrocardiographic Signals
Wolff-Parkinson-White (WPW) syndrome is a cardiac electrophysiology (EP) disorder caused by the presence of an accessory pathway (AP) that bypasses the atrioventricular node, faster ventricular activation rate, and provides a substrate for atrio-ventricular reentrant tachycardia (AVRT). Accurate localization of the AP is critical for planning and guiding catheter ablation procedures. While traditional diagnostic tree (DT) methods and more recent machine learning (ML) approaches have been proposed to predict AP location from surface electrocardiogram (ECG), they are often constrained by limited anatomical localization resolution, poor interpretability, and the use of small clinical datasets. In this study, we present a Deep Learning (DL) model for the localization of single manifest APs across 24 cardiac regions, trained on a large, physiologically realistic database of synthetic ECGs generated using a personalized virtual heart model. We also integrate eXplainable Artificial Intelligence (XAI) methods, Guided Backpropagation, Grad-CAM, and Guided Grad-CAM, into the pipeline. This enables interpretation of DL decision-making and addresses one of the main barriers to clinical adoption: lack of transparency in ML predictions. Our model achieves localization accuracy above 95%, with a sensitivity of 94.32% and specificity of 99.78%. XAI outputs are physiologically validated against known depolarization patterns, and a novel index is introduced to identify the most informative ECG leads for AP localization. Results highlight lead V2 as the most critical, followed by aVF, V1, and aVL. This work demonstrates the potential of combining cardiac digital twins with explainable DL to enable accurate, transparent, and non-invasive AP localization.
Open AccessExplainable Deep Learning-based Classification of Wolff-Parkinson-White Electrocardiographic Signals
Wolff-Parkinson-White (WPW) syndrome is a cardiac electrophysiology (EP) disorder caused by the presence of an accessory pathway (AP) that bypasses the atrioventricular node, faster ventricular activation rate, and provides a substrate for atrio-ventricular reentrant tachycardia (AVRT). Accurate localization of the AP is critical for planning and guiding catheter ablation procedures. While traditional diagnostic tree (DT) methods and more recent machine learning (ML) approaches have been proposed to predict AP location from surface electrocardiogram (ECG), they are often constrained by limited anatomical localization resolution, poor interpretability, and the use of small clinical datasets. In this study, we present a Deep Learning (DL) model for the localization of single manifest APs across 24 cardiac regions, trained on a large, physiologically realistic database of synthetic ECGs generated using a personalized virtual heart model. We also integrate eXplainable Artificial Intelligence (XAI) methods, Guided Backpropagation, Grad-CAM, and Guided Grad-CAM, into the pipeline. This enables interpretation of DL decision-making and addresses one of the main barriers to clinical adoption: lack of transparency in ML predictions. Our model achieves localization accuracy above 95%, with a sensitivity of 94.32% and specificity of 99.78%. XAI outputs are physiologically validated against known depolarization patterns, and a novel index is introduced to identify the most informative ECG leads for AP localization. Results highlight lead V2 as the most critical, followed by aVF, V1, and aVL. This work demonstrates the potential of combining cardiac digital twins with explainable DL to enable accurate, transparent, and non-invasive AP localization.
- Open Access
- 2 September 2025
- Published by Wiley in The Journal of Physiology
- Vol. 603 (19) , 5387-5413
Computational modelling of the impact of anatomical changes on ECGs in left ventricular hypertrophy
Left ventricular hypertrophy (LVH) is characterised by an increase in the mass and volume of the left ventricle, typically manifested as ventricular wall thickening and/or dilation. Due to its potential to cause severe, life‐threatening complications, ongoing research continues to explore its underlying mechanisms. This study aimed to determine how wall thickening and dilation specifically impact ECG waveforms, isolating these anatomical alterations without considering potential electrophysiological changes associated with LVH – a scenario achievable only through computational modelling. To accomplish this, eccentric and concentric cardiac models – with growth levels from 10% to 100% mass increase – were generated using a kinematic growth, finite element model derived from a healthy control model. Activation sequences were simulated for each model using a pseudo‐bidomain reaction‐eikonal approach, and 12‐lead ECGs were recorded from the hypertrophy models and compared to the control. Results indicated that activation patterns in eccentric hypertrophy models resembled the healthy model, while concentric hypertrophy models displayed substantial deviations. Both types of hypertrophy types led to prolonged QRS durations by up to 21 ms – a 40% increase from baseline – even in the absence of electrical remodelling. Eccentric hypertrophy increased amplitudes in precordial leads, minimally affecting limb leads, while concentric hypertrophy impacted all 12 leads with varied amplitude changes. Leads aVL, V1 and V5/V6 emerged as the most sensitive to anatomical changes. These findings could enhance the accuracy of LVH diagnosis using ECGs, offering a cost‐effective strategy to complement clinical evaluation and imaging, ultimately improving LVH detection and management. image Key points: Computational simulations revealed distinct effects of anatomical changes in eccentric and concentric left ventricular hypertrophy on 12‐lead ECG signals. Eccentric hypertrophy primarily affected the precordial leads, showing notable voltage amplitude increases across all precordial lead measurements. Concentric hypertrophy affected all 12 leads without a clear pattern of amplitude change, displaying both increases and decreases. Both eccentric and concentric hypertrophy resulted in a consistent prolongation of the QRS complex, showing up to 40% increase from baseline, even in the absence of electrophysiological remodelling. Leads aVL, III, V1 and V5/V6 were identified as the most sensitive to LVH, with computational results aligning well with independent clinical measurements.
- 2 September 2025
- Published by Wiley in The Journal of Physiology
- Vol. 603 (19) , 5387-5413
Open AccessComputational modelling of the impact of anatomical changes on ECGs in left ventricular hypertrophy
Left ventricular hypertrophy (LVH) is characterised by an increase in the mass and volume of the left ventricle, typically manifested as ventricular wall thickening and/or dilation. Due to its potential to cause severe, life‐threatening complications, ongoing research continues to explore its underlying mechanisms. This study aimed to determine how wall thickening and dilation specifically impact ECG waveforms, isolating these anatomical alterations without considering potential electrophysiological changes associated with LVH – a scenario achievable only through computational modelling. To accomplish this, eccentric and concentric cardiac models – with growth levels from 10% to 100% mass increase – were generated using a kinematic growth, finite element model derived from a healthy control model. Activation sequences were simulated for each model using a pseudo‐bidomain reaction‐eikonal approach, and 12‐lead ECGs were recorded from the hypertrophy models and compared to the control. Results indicated that activation patterns in eccentric hypertrophy models resembled the healthy model, while concentric hypertrophy models displayed substantial deviations. Both types of hypertrophy types led to prolonged QRS durations by up to 21 ms – a 40% increase from baseline – even in the absence of electrical remodelling. Eccentric hypertrophy increased amplitudes in precordial leads, minimally affecting limb leads, while concentric hypertrophy impacted all 12 leads with varied amplitude changes. Leads aVL, V1 and V5/V6 emerged as the most sensitive to anatomical changes. These findings could enhance the accuracy of LVH diagnosis using ECGs, offering a cost‐effective strategy to complement clinical evaluation and imaging, ultimately improving LVH detection and management. image Key points: Computational simulations revealed distinct effects of anatomical changes in eccentric and concentric left ventricular hypertrophy on 12‐lead ECG signals. Eccentric hypertrophy primarily affected the precordial leads, showing notable voltage amplitude increases across all precordial lead measurements. Concentric hypertrophy affected all 12 leads without a clear pattern of amplitude change, displaying both increases and decreases. Both eccentric and concentric hypertrophy resulted in a consistent prolongation of the QRS complex, showing up to 40% increase from baseline, even in the absence of electrophysiological remodelling. Leads aVL, III, V1 and V5/V6 were identified as the most sensitive to LVH, with computational results aligning well with independent clinical measurements.
- Open Access
- 22 August 2025
- Published by Wiley in The Journal of Physiology
- Vol. 604 (13) , 5440-5457
Uncertainty quantification via polynomial chaos expansion of myocardial fibre orientation and cardiac activation patterns
Predictive models and computational simulations of cardiac electrophysiology depend on precise anatomical representations, including the local myocardial fibre structure. However, obtaining patient‐specific fibre information is challenging. In addition, the influence of physiological variability in fibre orientation on cardiac activation simulations is poorly understood. We implemented rule‐based algorithms to generate fibres and robust uncertainty quantification methods to determine model output variability with respect to ventricular activation sequences. We used polynomial chaos, which reduces computational demands by using an emulator to approximate the underlying forward model. Our study examined activation sequences in response to nine stimuli and five metrics quantifying essential features of the activation sequence. The results indicated that the primary fibre orientation impacts the overall spread of activation, which could impact more complex patterns of activation; however, there is minimal impact on the location of discrete activation features, such as breakthrough sites. For free wall stimuli, the standard deviation (STD) was highest near the stimulus site, diminishing with distance. Apical stimuli showed complementary STD patterns, with epicardial pacing maximizing STD in the right basal area and endocardial pacing in the left. Ventricular junction stimuli exhibited symmetrical STD patterns, low near the stimulus but increasing sharply towards the apex, peaking on the left in the apical region. Furthermore, variability in the imbrication or helix angle did not impact the activation sequences. We conclude that in many relevant modelling contexts, the variability in myocardial fibre orientation can play an important role in the resulting activation sequences and should be accounted for. image Key points: The primary fibre orientation has modest impact on activation duration and location of most discrete activation features for ectopic stimuli, but introduces variability in activation sequence. For free wall stimuli, the standard deviation (STD) was highest on the stimulated surface, indicating that deviations are largest in early activation and diminish as activation reaches remote heart regions. For apical stimuli, activation patterns were insensitive to fibre orientation variations, but the STD had complementary maxima, strongly dependent on pacing surface. Epicardial pacing produced largest STD in right basal area while endocardial pacing affected left basal area, indicating strong dependence on fibre structure. For stimuli at both anterior and posterior ventricular junctions, STD patterns were symmetrical, with low values near stimulus and sharp increases towards apex, peaking in left apical region. The helical fibre orientation showed no relevant fluctuations in activation sequence.
- 22 August 2025
- Published by Wiley in The Journal of Physiology
- Vol. 604 (13) , 5440-5457
Open AccessUncertainty quantification via polynomial chaos expansion of myocardial fibre orientation and cardiac activation patterns
Predictive models and computational simulations of cardiac electrophysiology depend on precise anatomical representations, including the local myocardial fibre structure. However, obtaining patient‐specific fibre information is challenging. In addition, the influence of physiological variability in fibre orientation on cardiac activation simulations is poorly understood. We implemented rule‐based algorithms to generate fibres and robust uncertainty quantification methods to determine model output variability with respect to ventricular activation sequences. We used polynomial chaos, which reduces computational demands by using an emulator to approximate the underlying forward model. Our study examined activation sequences in response to nine stimuli and five metrics quantifying essential features of the activation sequence. The results indicated that the primary fibre orientation impacts the overall spread of activation, which could impact more complex patterns of activation; however, there is minimal impact on the location of discrete activation features, such as breakthrough sites. For free wall stimuli, the standard deviation (STD) was highest near the stimulus site, diminishing with distance. Apical stimuli showed complementary STD patterns, with epicardial pacing maximizing STD in the right basal area and endocardial pacing in the left. Ventricular junction stimuli exhibited symmetrical STD patterns, low near the stimulus but increasing sharply towards the apex, peaking on the left in the apical region. Furthermore, variability in the imbrication or helix angle did not impact the activation sequences. We conclude that in many relevant modelling contexts, the variability in myocardial fibre orientation can play an important role in the resulting activation sequences and should be accounted for. image Key points: The primary fibre orientation has modest impact on activation duration and location of most discrete activation features for ectopic stimuli, but introduces variability in activation sequence. For free wall stimuli, the standard deviation (STD) was highest on the stimulated surface, indicating that deviations are largest in early activation and diminish as activation reaches remote heart regions. For apical stimuli, activation patterns were insensitive to fibre orientation variations, but the STD had complementary maxima, strongly dependent on pacing surface. Epicardial pacing produced largest STD in right basal area while endocardial pacing affected left basal area, indicating strong dependence on fibre structure. For stimuli at both anterior and posterior ventricular junctions, STD patterns were symmetrical, with low values near stimulus and sharp increases towards apex, peaking in left apical region. The helical fibre orientation showed no relevant fluctuations in activation sequence.
- Open Access
- 3 June 2025
- Published by Elsevier in Medical Image Analysis
- Vol. 105, 103641
Accurate and efficient cardiac digital twin from surface ECGs: Insights into identifiability of ventricular conduction system
- 3 June 2025
- Published by Elsevier in Medical Image Analysis
- Vol. 105, 103641
Open AccessAccurate and efficient cardiac digital twin from surface ECGs: Insights into identifiability of ventricular conduction system
- Open Access
- 7 May 2025
- Published by Oxford University Press (OUP) in EP Europace
- Vol. 27 (5)
An in silico guide for ventriculo-ventricular delay programming for left bundle branch-optimized cardiac resynchronization therapy
Aims: Left bundle branch pacing (LBBP)–optimized cardiac resynchronization therapy (LOT-CRT) can improve left ventricular (LV) activation when LBBP alone or conventional biventricular pacing are ineffective. However, the optimal programming settings for ventriculo-ventricular delay (VVD) for LOT-CRT are unknown. We aim to investigate how to optimally program VVD for LOT-CRT in the presence of various LV conduction substrates using computational modelling. Methods and results: We simulated ventricular activation on 24 anatomies and validated the model against clinical data. Diffuse LV conduction system and intra-myocardial delay were simulated by slowing the conduction velocity of the LV His-Purkinje system and myocardium, respectively, alone or in combination with proximal left bundle branch block (LBBB). We simulated LOT-CRT with selective or myocardial capture (LV septal pacing, LVSP) with VVD ranging between −100 ms (LBBP/LVSP ahead) and +100 ms [LV epicardial lead (LVepiP), ahead]. Response was quantified with 95% LV activation times (LVAT95). In the presence of diffuse LV conduction system delay, the optimal VVD for LOT-CRT was always negative (LBBB: −42.5 ± 6.6 ms; no LBBB: −36.2 ± 5.6 ms), as delivering LBBP ahead of LVepiP compensates for the slow LV His-Purkinje. In the presence of LV intra-myocardial disease, the shortest LVAT95 with LOT-CRT was achieved by pacing the coronary sinus LV first (optimal VVD for LBBB: 23.3 ± 8.5 ms; no LBBB: 79.2 ± 18.0 ms). The type of capture for LOT-CRT affected the optimal VVD, with myocardial capture favouring negative VVDs (LVSP ahead). Conclusion: The optimal VVD for LOT-CRT depends on the mechanism of delayed LV activation and type of capture achieved, highlighting the importance of VVD optimization.
- 7 May 2025
- Published by Oxford University Press (OUP) in EP Europace
- Vol. 27 (5)
Open AccessAn in silico guide for ventriculo-ventricular delay programming for left bundle branch-optimized cardiac resynchronization therapy
Aims: Left bundle branch pacing (LBBP)–optimized cardiac resynchronization therapy (LOT-CRT) can improve left ventricular (LV) activation when LBBP alone or conventional biventricular pacing are ineffective. However, the optimal programming settings for ventriculo-ventricular delay (VVD) for LOT-CRT are unknown. We aim to investigate how to optimally program VVD for LOT-CRT in the presence of various LV conduction substrates using computational modelling. Methods and results: We simulated ventricular activation on 24 anatomies and validated the model against clinical data. Diffuse LV conduction system and intra-myocardial delay were simulated by slowing the conduction velocity of the LV His-Purkinje system and myocardium, respectively, alone or in combination with proximal left bundle branch block (LBBB). We simulated LOT-CRT with selective or myocardial capture (LV septal pacing, LVSP) with VVD ranging between −100 ms (LBBP/LVSP ahead) and +100 ms [LV epicardial lead (LVepiP), ahead]. Response was quantified with 95% LV activation times (LVAT95). In the presence of diffuse LV conduction system delay, the optimal VVD for LOT-CRT was always negative (LBBB: −42.5 ± 6.6 ms; no LBBB: −36.2 ± 5.6 ms), as delivering LBBP ahead of LVepiP compensates for the slow LV His-Purkinje. In the presence of LV intra-myocardial disease, the shortest LVAT95 with LOT-CRT was achieved by pacing the coronary sinus LV first (optimal VVD for LBBB: 23.3 ± 8.5 ms; no LBBB: 79.2 ± 18.0 ms). The type of capture for LOT-CRT affected the optimal VVD, with myocardial capture favouring negative VVDs (LVSP ahead). Conclusion: The optimal VVD for LOT-CRT depends on the mechanism of delayed LV activation and type of capture achieved, highlighting the importance of VVD optimization.
- Open Access
- 8 March 2025
- Published by Elsevier in Computers in Biology and Medicine
- Vol. 189, 109930
Quantifying anatomically-based in-silico electrocardiogram variability for cardiac digital twins
- 8 March 2025
- Published by Elsevier in Computers in Biology and Medicine
- Vol. 189, 109930
Open AccessQuantifying anatomically-based in-silico electrocardiogram variability for cardiac digital twins
- 28 September 2024
- Published by Elsevier in Computers in Biology and Medicine
- Vol. 182, 109201
BOATMAP: Bayesian Optimization Active Targeting for Monomorphic Arrhythmia Pace-mapping
- 28 September 2024
- Published by Elsevier in Computers in Biology and Medicine
- Vol. 182, 109201
BOATMAP: Bayesian Optimization Active Targeting for Monomorphic Arrhythmia Pace-mapping
- Open Access
- 11 September 2024
- Published by Oxford University Press (OUP) in EP Europace
- Vol. 27 (2)
A computational study on the influence of antegrade accessory pathway location on the 12-lead electrocardiogram in Wolff–Parkinson–White syndrome
Aims: Wolff–Parkinson–White (WPW) syndrome is a cardiovascular disease characterized by abnormal atrioventricular conduction facilitated by accessory pathways (APs). Invasive catheter ablation of the AP represents the primary treatment modality. Accurate localization of APs is crucial for successful ablation outcomes, but current diagnostic algorithms based on the 12-lead electrocardiogram (ECG) often struggle with precise determination of AP locations. In order to gain insight into the mechanisms underlying localization failures observed in current diagnostic algorithms, we employ a virtual cardiac model to elucidate the relationship between AP location and ECG morphology. Methods and results: We first introduce a cardiac model of electrophysiology that was specifically tailored to represent antegrade APs in the form of a short atrioventricular bypass tract. Locations of antegrade APs were then automatically swept across both ventricles in the virtual model to generate a synthetic ECG database consisting of 9271 signals. Regional grouping of antegrade APs revealed overarching morphological patterns originating from diverse cardiac regions. We then applied variance-based sensitivity analysis relying on polynomial chaos expansion on the ECG database to mathematically quantify how variation in AP location and timing relates to morphological variation in the 12-lead ECG. We utilized our mechanistic virtual model to showcase the limitations of AP localization using standard ECG-based algorithms and provide mechanistic explanations through exemplary simulations. Conclusion: Our findings highlight the potential of virtual models of cardiac electrophysiology not only to deepen our understanding of the underlying mechanisms of WPW syndrome but also to potentially enhance the diagnostic accuracy of ECG-based algorithms and facilitate personalized treatment planning.
- 11 September 2024
- Published by Oxford University Press (OUP) in EP Europace
- Vol. 27 (2)
Open AccessA computational study on the influence of antegrade accessory pathway location on the 12-lead electrocardiogram in Wolff–Parkinson–White syndrome
Aims: Wolff–Parkinson–White (WPW) syndrome is a cardiovascular disease characterized by abnormal atrioventricular conduction facilitated by accessory pathways (APs). Invasive catheter ablation of the AP represents the primary treatment modality. Accurate localization of APs is crucial for successful ablation outcomes, but current diagnostic algorithms based on the 12-lead electrocardiogram (ECG) often struggle with precise determination of AP locations. In order to gain insight into the mechanisms underlying localization failures observed in current diagnostic algorithms, we employ a virtual cardiac model to elucidate the relationship between AP location and ECG morphology. Methods and results: We first introduce a cardiac model of electrophysiology that was specifically tailored to represent antegrade APs in the form of a short atrioventricular bypass tract. Locations of antegrade APs were then automatically swept across both ventricles in the virtual model to generate a synthetic ECG database consisting of 9271 signals. Regional grouping of antegrade APs revealed overarching morphological patterns originating from diverse cardiac regions. We then applied variance-based sensitivity analysis relying on polynomial chaos expansion on the ECG database to mathematically quantify how variation in AP location and timing relates to morphological variation in the 12-lead ECG. We utilized our mechanistic virtual model to showcase the limitations of AP localization using standard ECG-based algorithms and provide mechanistic explanations through exemplary simulations. Conclusion: Our findings highlight the potential of virtual models of cardiac electrophysiology not only to deepen our understanding of the underlying mechanisms of WPW syndrome but also to potentially enhance the diagnostic accuracy of ECG-based algorithms and facilitate personalized treatment planning.
- Open Access
Quantifying variabilities in cardiac digital twin models of the electrocardiogram
Cardiac digital twins (CDTs) of human cardiac electrophysiology (EP) are digital replicas of patient hearts that match like-for-like clinical observations. The electrocardiogram (ECG), as the most prevalent non-invasive observation of cardiac electrophysiology, is considered an ideal target for CDT calibration. Recent advanced CDT calibration methods have demonstrated their ability to minimize discrepancies between simulated and measured ECG signals, effectively replicating all key morphological features relevant to diagnostics. However, due to the inherent nature of clinical data acquisition and CDT model generation pipelines, discrepancies inevitably arise between the real physical electrophysiology in a patient and the simulated virtual electrophysiology in a CDT. In this study, we aim to qualitatively and quantitatively analyze the impact of these uncertainties on ECG morphology and diagnostic markers. We analyze residual beat-to-beat variability in ECG recordings obtained from healthy subjects and patients. Using a biophysically detailed and anatomically accurate computational model of whole-heart electrophysiology combined with a detailed torso model calibrated to closely replicate measured ECG signals, we vary anatomical factors (heart location, orientation, size), heterogeneity in electrical conductivities in the heart and torso, and electrode placements across ECG leads to assess their qualitative impact on ECG morphology. Our study demonstrates that diagnostically relevant ECG features and overall morphology appear relatively robust against the investigated uncertainties. This resilience is consistent with the narrow distribution of ECG due to residual beat-to-beat variability observed in both healthy subjects and patients.
Open AccessQuantifying variabilities in cardiac digital twin models of the electrocardiogram
Cardiac digital twins (CDTs) of human cardiac electrophysiology (EP) are digital replicas of patient hearts that match like-for-like clinical observations. The electrocardiogram (ECG), as the most prevalent non-invasive observation of cardiac electrophysiology, is considered an ideal target for CDT calibration. Recent advanced CDT calibration methods have demonstrated their ability to minimize discrepancies between simulated and measured ECG signals, effectively replicating all key morphological features relevant to diagnostics. However, due to the inherent nature of clinical data acquisition and CDT model generation pipelines, discrepancies inevitably arise between the real physical electrophysiology in a patient and the simulated virtual electrophysiology in a CDT. In this study, we aim to qualitatively and quantitatively analyze the impact of these uncertainties on ECG morphology and diagnostic markers. We analyze residual beat-to-beat variability in ECG recordings obtained from healthy subjects and patients. Using a biophysically detailed and anatomically accurate computational model of whole-heart electrophysiology combined with a detailed torso model calibrated to closely replicate measured ECG signals, we vary anatomical factors (heart location, orientation, size), heterogeneity in electrical conductivities in the heart and torso, and electrode placements across ECG leads to assess their qualitative impact on ECG morphology. Our study demonstrates that diagnostically relevant ECG features and overall morphology appear relatively robust against the investigated uncertainties. This resilience is consistent with the narrow distribution of ECG due to residual beat-to-beat variability observed in both healthy subjects and patients.