Background: The Electrocardiographic Imaging (ECGI) inverse problem is ill-posed. Multivariate Adaptive Regression Splines (MARS) provide a data-driven alternative to traditional regularization, yet become computationally expensive on high-resolution data. This study proposes an adaptive strategy that dynamically scales MARS model complexity.
Methods: Clinical data from 10 patients with premature ventricular contractions (PVCs) were analyzed. Training data were simulated on patient geometries. Parametric sweeps assessed key MARS hyperparameters (threshold-epsilon and term limit-L_max). An adaptive L_max formulation was developed using a patient-specific saturation model based on spatiotemporal volume (V).
Results: A high-complexity baseline model (L_max=128, epsilon=0.001) yielded accurate results (LE: 17.2 mm) but required impractical runtimes (43.4 hours). The proposed adaptive strategy outperformed static constraints, reducing the mean localization error (LE) in the optimization set from 18.85 mm to 15.99 mm.
Conclusions: Directly constraining L_max provides more reliable control over model complexity than threshold-based stopping. Scaling L_max with patient-specific data volume reduces overfitting while preserving physiological signals, improving the feasibility of MARS-based ECGI for pre-procedural planning.