Predicting calcification risk in prosthetic aortic valves: a hybrid physics-based and machine learning approach

Calcific degeneration remains the primary failure mode of bioprosthetic aortic valves and a growing concern for emerging polymeric alternatives. In this study, six prosthetic valve configurations combining two leaflet geometries (V-shaped scallop D1^V and U-shaped scallop D2^U) with three constitutive material models (synthetic elastomer, bovine pericardium and porcine tissue) are simulated throughout systole within a patient-specific curved aorta reconstructed from clinical CT data using a high-fidelity fluid-structure interaction framework. An incremental formulation of Finite-Time Lyapunov Exponents (FTLE) applied to the leaflet deformation gradients is introduced as a novel descriptor of structural point coherence on the leaflet surface. Together with wall shear stress (WSS) derived features, the FTLE field is fed into an unsupervised k-means clustering algorithm that partitions each leaflet into four calcification risk classes without any fitting to experimental data. Validation against an experimentally calibrated calcification intensity map from micro-CT of explanted bovine pericardial valves yields a Spearman rank correlation of 0.98 and a Cohen's weighted kappa of 0.91 when the FTLE- and WSS-based classifications are combined, with an approximately equal optimal weighting.

The analysis reveals that porcine tissue concentrates strain into localised high-gradient zones whereas the isotropic elastomer distributes strain uniformly yet generates elevated shear fluctuations through sustained leaflet flutter. The D2^U leaflet geometry paired with bovine pericardium emerges as the least susceptible configuration across both risk dimensions. By coupling leaflet structural point coherence with haemodynamic shear in a model-free framework, the proposed methodology provides a transferable design tool for ranking and evaluating next-generation prosthetic valve candidates with respect to calcification susceptibility.