Toward Individualized Surgical Parameter Selection via Numerical Simulations and Bayesian Statistics

Abstract

Restoring natural movement and sensation after limb loss remains a central challenge in modern prosthetics. Beyond orthopedic changes, amputation disrupts the natural interaction between opposing muscles, which normally supports movement and proprioceptive feedback. The agonist-antagonist myoneural interface (AMI) surgically restores this coupling by connecting opposing muscles. However, the antagonist prestretch, which influences the resulting range of motion (ROM), is currently selected heuristically because clinical data is scarce and high-fidelity simulations are computationally expensive. This study presents an initial interdisciplinary approach toward addressing this gap. We first generate data using multiscale muscle-tendon simulations, then apply Bayesian Optimization (BO) to identify informative sampling points for estimating the optimal prestretch. A near-optimal prestretch is found within a few forward evaluations, revealing an interior maximum beyond which further prestretch reduces ROM. Finally, Bayesian Data Augmentation (DA) infers the underlying data distributions, aiming toward patient-specific, uncertainty-aware predictions of ROM. In essence, our study serves as a computational framework demonstrating a route toward overcoming the data sparsity and hardware limitations that currently obstruct individualized AMI planning.