Prediction of Medial Tibiofemoral Joint Reaction Force Using Custom Instrumented Insoles and Neural Networks for Walking and Running Tasks.

Abstract

Medial tibiofemoral joint reaction force is a clinically relevant variable for knee osteoarthritis progression and can be estimated using complex musculoskeletal models. Musculoskeletal model estimation of this variable is time-consuming, expensive, requires trained researchers, and is restricted to lab settings. We aimed to simplify the measurement of the medial knee joint contact force during walking and running using custom instrumented insoles and deep learning methods. Motion capture, force plate, and insoles instrumented with triaxial piezoresistive force sensors recorded data while 9 young healthy female individuals walked and ran at varying speeds. Two task-specific convolutional neural networks were developed for walking and running using piezoresistive force sensors as inputs during the stance phase. Results showed that both models were able to estimate total medial joint contact force with strong correlation coefficients (r > .98) and moderate mean absolute error (<0.36 body weight). These methods show the possibility of collecting medial knee joint contact force during walking and running in a clinical setting. Future research with this framework can be used to provide biofeedback to reduce medial knee joint contact force in high-risk knee osteoarthritis groups in clinical settings and daily life.