Self-organization of conducting pathways explains complex wave trajectories in procedurally interpolated fibrotic cardiac tissue: a digital-twin study

In precision cardiology, digital twinning technology (DT) holds promise for predicting arrhythmias by leveraging patient data and biophysics knowledge. However, current DTs are designed to directly reproduce biopotential conduction in cardiac tissue, while only indirect non-invasive methods can be clinically implemented on real organs. This discrepancy challenges our understanding of DT applicability limits. This study aims to enhance DT by developing an in-vitro training complement. We conducted a frame-by-frame comparison of in-vitro optical mapping of biopotential conduction with machine learning (ML) optimized DT predictions. Patient-specific self-organized tissue samples of human induced pluripotent stem cells-derived cardiomyocytes (CMs) with diffuse fibrosis served as DT prototypes. High spatiotemporal resolution optical mapping recordings (Δx=117 ± 4 μm, Δt=7.69 ms) and immunostainings were used to reproduce fibrotic samples with a linear size of 7.5 mm. Using data-driven ML-optimization of the Cellular Potts model, we examined wave propagation at the subcellular level. The modified Glazier-Graner-Hogeweg model accurately reflected the “perinatal window” until the 20th day of differentiation, affecting CMs self-organization. The percolation threshold of virtual conductive pathways reached 26% (26.7 ± 2.9% of CMs in-vitro), resulting in a spatial correlation of amplitude maps between prototype samples and their DT with Pearson’s coefficients of 0.83 ± 0.02. As a proof-of-concept, we demonstrated the ability of ML-optimized DT to predict and interpolate wavefront trajectories in optical mapping recordings. We found that mathematical approximation of fibrosis distribution played a key role in DT prediction accuracy, potentially informing the implementation of LGE-MRI detection of fibrosis within cardiac DT frameworks.Dataset A: Immunostaining images were sorted based on the day of enzymatic disaggregation (before and after day 20). We collected and sorted α-actinin, Connexin43 and DAPI immunostainings for Cellular Potts Model optimization. During data processing, cell shapes (n=109 and n=69 for CM and BPs respectively after day 20, n=209 and n=90 for CM and BPs respectively before day 20) were formalized.Dataset C corresponds to FluoVolt recordings (3 samples). Dataset B corresponds to Fluo-4 AM recordings in iPSC-CMs samples with diffuse fibrosis imitation (4 samples).