We present a workflow to create (patho)physiological, synthetic cardiac cohorts that serve as population-level digital twins. Specifically, we focus on modeling biventricular anatomy and depolarization in healthy subjects and subjects with bundle branch block (BBB), using a reaction-eikonal model coupled with a lead-field approach for ECG simulation.
To capture population-wide anatomical variability, we constructed a statistical shape model based on patient-specific biventricular meshes derived from cardiac CT scans of 630 subjects (280 healthy subjects and 350 subjects with BBB). The first 133 principal components of this model captured 99% of the anatomical variability. By sampling random variations across these 133 components, we generated a synthetic cohort of 800 biventricular meshes. These ventricles were co-registered into an average human torso mesh, yielding a realistic anatomical cohort free of privacy concerns and suitable for large-scale ECG simulations.
Healthy baseline electrophysiology was defined for three synthetic hearts by optimizing electrophysiological parameters within physiological bounds until the simulated 12-lead ECG met predefined criteria for normal ventricular depolarization morphology, including appropriate lead-wise QRS deflections and precordial R-wave progression. The resulting model-specific baseline parameter sets were then perturbed to reproduce bundle branch block phenotypes, allowing pathological activation patterns to be generated as controlled deviations from individually optimized healthy states. Further expansions to the larger cohort and quantification of the parameter perturbations reproducing clinical ECG characteristics of distinct conduction pathologies (left and right BBB, left anterior fascicular block, left posterior fascicular block, and bifascicular BBB) are ongoing.
While patient-specific modeling is inherently limited by data availability and clinical noise, this synthetic approach allows for the systematic, continuous exploration of electrophysiological parameter spaces. This workflow thereby establishes a comprehensive foundation for generating the large-scale, highly controlled digital twin libraries crucial for advanced conduction research.