Introduction. Cardiac digital twins (CDTs) typically require cross-sectional imaging or invasive recordings. We propose a non-invasive genetic algorithm (GA) guided by a spectral cost function to generate CDTs directly from body surface potential maps (BSPM) with non-invasive data.
Methods. The framework is a GA that combines an isochron-based forward model with a developed spectral cost function based on Laplace-Beltrami modes to identify activation sources. Validation included a standalone cost function assessment on 10 synthetic and 2 clinical ectopic cases; and an end-to-end GA evaluation on multi-source synthetic configurations and one clinical sinus rhythm patient (7 focal sources).
Results. Ectopic origins were correctly identified in all clinical cases. In-silico costs increased monotonically with distance from the focus. For the in-silico GA evaluation, GA maintained high accuracy across multi-source configurations. For the clinical sinus rhythm patient, reconstruction yielded a mean BSPM correlation of 0.85 ± 0.37, as illustrated in the attached figure.
Conclusions. This imageless framework accurately reconstructs complex activation patterns from BSPM, offering a non-invasive pathway for patient-specific CDT personalization in clinical settings.