Digital twins have emerged as powerful tools for simulating patient-specific cardiac electrophysiology, enabling non-invasive assessment of electrical activation patterns and supporting clinical decision-making. This study addresses conduction disorders—specifically left bundle branch block (LBBB)—by constructing digital twins derived from cardiac MRI and 12-lead ECG data from electrophysiological studies (EPS).
Patient-specific biventricular geometries were reconstructed from short-axis 2D cardiac MRI scans using Segment, converted into finite element meshes via Gmsh, and processed through an automated anatomical pipeline. Electrical activation was simulated using a reduced-order Eikonal model driven by root nodes—early activation sites—personalized through Sequential Monte Carlo Approximate Bayesian Computation (SMC-ABC). For the sinus rhythm simulations, a Purkinje network was used to supply realistic pathways.
Two different LBBB patients integrated the clinical data. For the first patient, the model correctly inferred no left ventricular activation for sinus rhythm, with root nodes distributed across the right ventricle. As for the apex-stimulation, activation was correctly localized to the apical region. Simulated ECGs matched the clinical readings, with average Pearson Correlation Coefficients (PCC) of 0.95 and 0.77 for sinus and apex pacing, respectively. In the second patient, activation under right ventricular outflow tract stimulation was accurately captured (PCC = 0.79). The sinus rhythm inference presented good correspondece of the ECG morphology (PCC of 0.92). However, there were some root nodes in the LV. A local activation time (LAT) analysis revealead that the LV nodes were activated later than RV ones, which agrees with LBBB diagnosis.
Importantly, studies validating digital twin pipelines with real clinical data remain scarce, despite being essential for translating these technologies into medical practice. Our results help reducing this gap by providing evidence of model performance under real-world conditions. This lays the foundation for the clinical translation of digital twin in cardiology, offering a scalable, data-driven approach to personalized diagnostics and therapy planning.