Development and Validation of Remote Motion Assessment Tool Enhanced by Personalized Musculoskeletal Modeling for Prediction of Joint Kinematics and Kinetics During Gait

Abstract

This thesis presents a novel approach to gait assessment by leveraging artificial intelligence (AI) algorithms to overcome the limitations of traditional gait analysis methods. Conventional motion capture systems, while accurate, are costly, cumbersome, and impractical for widespread clinical use. To address these challenges, we develop and evaluate deep learning models that predict ground reaction forces (GRFs) and joint kinetics directly from kinematic data, eliminating the need for force plates or extensive sensor setups. We introduce Seq2Seq deep learning models and Physics-Informed Neural Networks (PiNNs) to estimate GRFs during gait with high accuracy. The Seq2Seq model captures temporal dependencies in gait cycles, while the PiNN integrates biomechanical constraints, ensuring physically plausible predictions. Extensive training and validation are performed across multiple datasets using leave-one-subject-out (LOSO) cross-validation, demonstrating superior generalization compared to conventional data-driven approaches. Additionally, we explore the use of computer vision-based pose estimation algorithms, including BlazePose and AlphaPose, to extract gait kinematics from standard RGB video feeds. These algorithms when integrated into an end-to-end AI pipeline that processes raw video data, estimates joint trajectories, and predicts GRFs and joint moments, enabling Markerless, real-time gait analysis. The findings show that our AI-driven frameworks achieve comparable accuracy to laboratory-based motion capture while offering scalability, portability, and cost-efficiency. This research contributes to the transition toward mobile, AI-enhanced gait assessment, with potential applications in rehabilitation, remote patient monitoring, and clinical diagnostics. Future work will refine these models for real-time deployment, adaptive learning, and personalized gait interventions. By integrating AI with biomechanics, this thesis advances the field of automated, data-driven gait analysis, paving the way for accessible and intelligent clinical tools that enhance patient outcomes and mobility research.

Date of Award	2025
Original language	American English
Supervisor	Marwan El Rich (Supervisor)