Simulation-based digital twinning of activation and repolarisation sequences from the ECG across healthy and diseased hearts

Abnormal patterns of ventricular repolarisation are thought to contribute to lethal arrhythmias in various cardiac conditions, including inherited and acquired channelopathies, cardiomyopathies, and ischaemic heart disease. However, methods to detect these repolarisation abnormalities are limited. In this study, we introduce and assess a novel simulation-based method to infer ventricular activation and repolarisation times from the 12-lead electrocardiogram (ECG) and magnetic resonance-derived ventricular anatomical reconstruction, applicable for the first time to both healthy controls and cases with abnormal repolarisation. First, ventricular activation times were reconstructed through iterative refinement of early activation sites and conduction velocities, until the model and target QRS complexes matched. Then, ventricular repolarisation times were reconstructed through iterative refinement of ventricular action potential durations and an action potential shape parameter until the model and target T waves matched, including regularisation. Repolarisation inference was evaluated against 18 benchmark simulations with known repolarisation times, including both control and hypertrophic cardiomyopathy (HCM) cases with abnormal repolarisation. Inferred repolarisation times showed good agreement with the ground truth in control and HCM (Spearman r=0.63±0.11 and 0.65±0.19, respectively), with the inferred model T waves closely matching the target T waves (r=0.81±0.05 and 0.78±0.08, respectively). The method further demonstrated flexibility in reconstructing the macroscopic patterns of delayed repolarisation across a range of abnormal ventricular repolarisation sequences, demonstrating applicability to a wide range of pathological scenarios. Simulation-based inference can accurately reconstruct repolarisation times from the 12-lead ECG in cases with both normal and abnormal repolarisation patterns.