Estimation of right ventricular fiber orientation and fibrotic tissue from a single electroanatomical map

Accurate characterization of cardiac fiber orientation and detection of myocardial scars are crucial for developing personalized cardiac models that support precision diagnosis and treatment planning for heart diseases. However, measuring right ventricular (RV) fiber orientations remains challenging in clinical settings, resulting in a lack of patient-specific fiber information. In this study, we aim to assess the feasibility of reconstructing patient-specific RV fiber orientations and detecting fibrosis using available electroanatomical data by estimating tissue conductivity parameters.

We adopted an ensemble-based physics-informed neural network, Δ-FiberNet, to solve the inverse anisotropic Eikonal equation. The ensemble architecture enables the quantification of uncertainty in the estimated fiber field. We evaluated the model performance under both physiological and pathological conditions, including regions with slowed conduction or completely non-conductive tissue. The predicted conductivity parameters were then employed to identify fibrotic tissue.

The results indicate that the model reliably learns from a single activation map, even in the presence of regions with abnormal conductivity. The robustness analysis results further confirmed the model's performance under various levels of measurement noise and sampling densities.

This proof-of-concept study demonstrates Δ-FiberNet's ability to infer RV fiber orientations from a single activation map and provides initial evidence for its potential in identifying fibrotic tissue, which requires further validation.