Body surface potential driven personalisation of electrophysiological digital twins in hypertrophic cardiomyopathy

Figures

• 

  Fig 1. Workflow for developing EP digital twins. (I) Multimodal clinical data, including thoracic CT and CMR are segmented and co-registered to construct patient-specific finite-element meshes. (II) Ventricular EP and BSPs are simulated using a RE framework. (III) Model parameters are iteratively calibrated against clinically recorded BSPs using emulator-based HM, yielding electrophysiological digital twins. The proposed pipeline is an extension of our previous work [20].

• 

  Fig 2. Registration validation metrics across the cohort. Box plots show (A) Dice similarity coefficient, (B) HD95 (mm), (C) centroid offset (mm), and (D) volume difference (%). Boxes represent the interquartile range (IQR), horizontal lines indicate median values, and whiskers extend to 1.5 × IQR.

• 

  Fig 3. Two-stage parameter screening and calibration strategy for ventricular depolarisation and repolarisation. Model parameters are partitioned into fixed parameters (pcons), depolarisation-related parameters (pQRS), and repolarisation-related parameters (pT). (I) Initial parameter screening identifies subsets of influential parameter subsets for each phase of the cardiac cycle. (II) Calibration is performed sequentially: first calibrating pQRS keeping pT fixed to their baseline values (pT,0), then calibrating pT while pQRS are fixed to their calibrated values (pQRS,c). Gaussian process emulators and Bayesian history matching iteratively rule out implausible regions at each stage, yielding patient-specific parameter sets.

• 

  Fig 4. Representative activation and repolarisation maps. (Left) Patient-specific HPS configuration, with red markers indicating early activation sites. (Centre) Activation map ranging from 0 to 120 ms. (Right) Repolarisation map ranging from 150 to 450 ms.

• 

  Fig 5. Calibration performance on representative and full vest electrode sets. (A) Distribution across the cohort of the percentage of 50 representative electrodes achieving moderate-to-good (PCC≥0.6, left) and good (PCC≥0.8, right) morphological agreement between simulated and clinical BSPs. (B) Per-patient breakdown of agreement across representative electrodes, partitioned into good (PCC≥0.8), moderate (0.8>PCC≥0.6), and poor (PCC < 0.6) categories. (C) Corresponding cohort-level performance across all 252 ECGI vest electrodes using the same agreement thresholds. (D) Per-patient agreement across the full ECGI vest. Grey bars indicate electrodes excluded from analysis due to missing recordings or traces lacking identifiable QRS-T morphology.

• 

  Fig 6. Spatial distribution of electrode-wise agreement between simulated and clinical body surface potentials. Cohort-level median Pearson correlation coefficients (PCCs) are projected onto a template ECGI vest geometry for four anatomical views: (A) right lateral, (B) anterior, (C) left lateral, and (D) posterior. Colour bar shows distribution of only PCC > 0.6 as no contiguous regions of poor agreement (PCC < 0.6) were observed.

• 

  Fig 7. Representative higher- and lower-performing model calibration cases across the ECGI vest. Aligned clinical (black) and simulated (red) BSP waveforms are shown for their 50 sampled electrodes. Panel (A) shows a higher-performing patient and panel (B) a lower-performing patient, selected from the upper and lower ends of the cohort based on percentage agreement. All traces are displayed over an identical time window with consistent amplitude scaling.

• 

  Fig 8. Pacing-induced changes in ECG-derived depolarisation and repolarisation metrics for an external clinical cohort and calibrated digital twins. Comparison of pacing-induced changes (Δ = pacing - sinus) between an independent external HCM validation cohort (blue, n = 13) and calibrated digital twins (red, n = 17). (A) Global QRS duration changes (ΔQRSd) shown as box-and-whisker plots. (B-F) Lead-specific changes across precordial leads V1-V8 displayed as median with IQR error bars: (B) ventricular activation time (ΔVd), (C) QRS temporal centroid (ΔQRSmax,center), (D) activation-recovery interval (ΔARI), (E) T-wave onset (ΔTon), and (F) T-wave offset (ΔToff).

• 

  Fig 9. Correlation between calibrated electrophysiological parameters and clinical demographics. Manhattan plots summarising associations between patient-specific calibrated EP parameters and continuous (top) and categorical (bottom) clinical variables. Parameters are grouped by QRS-related vs repolarisation-related, separated by the vertical divider. The red dashed line indicates nominal significance (p = 0.05).