Automated Assessment of Musculoskeletal Injury Rehabilitation Using Underfoot Gait Patterns and Deep Learning Algorithms

Gait rehabilitation is crucial for musculoskeletal injury recovery, as loss of mobility can accelerate health decline. However, accurately characterizing rehabilitation progress is challenging. In this work, a deep learning approach that tracks recovery by analyzing footstep ground reaction force (GRF) and center of pressure (COP) time series is proposed. A deep learning-based convolutional autoencoder (CAE) with a mostly-deterministic convolutional kernel transform (MiniRocket) was trained using 561 subjects to model healthy gait, with CAE reconstruction error indicating rehabilitation progress in injured individuals ( \(N=1826\) ). The results demonstrate sensitivity to changes in gait patterns in four classes of injuries (hip, knee, calcaneus, and ankle), with average positive improvements of 3.9%, 6.6%, 4.8% and 6.7%, detected in each class, respectively. In addition, the reconstruction error at discharge and the changes in the reconstruction error over time were found to be significant predictors of readmission ( \(p=0.006\) and \(p=0.03\) , respectively). The proposed technique could support clinical treatment decisions and improve patient outcomes.