Privacy-Preserving Human Motion Analysis for Lower Back Pain Stratification through Federated Learning

 

Human Gait Analysis is crucial in healthcare applications, with nu-

merous research works focusing on machine learning and deep learning ap-

proaches for tasks such as abnormal gait detection and gait quality assessment.

However, developing such models requires collecting and sharing a significant

amount of patient data, raising privacy concerns. In this study, we introduce the

world’s first technique for constructing a deep neural network model to stratify

patients’ pain levels based on video recordings of timed up-and-go activities,

while ensuring privacy preservation through modern federated learning algo-

rithms. Our experimental results demonstrate the effectiveness of this technique

in accurately stratifying LBP levels without the need for data sharing among local

clients to maintain privacy.