A Unified Framework for Neurological Disease Detection and Gait Classification Using Deep Graph Learning

Abstract

This paper presents Dynamic Gait Signature Analysis (DGSA), an innovative approach to gait analysis that leverages deep graph learning techniques. Unlike conventional methods, DGSA leverages multifaceted parameters and advanced deep graph learning techniques, such as Graph Convolutional Networks (GCNs) and Graph Attention Networks (GATs). These techniques enable a comprehensive analysis of gait dynamics, including the use of dynamic graph representation methods like Gait Cycle Joint Angles Graph and Gait Cycle Joint Power Graph. DGSA's unique framework allows for simultaneous prediction of neurological diseases, gait classification, and early detection of cognitive impairments. By modeling gait as dynamic graph structures, DGSA captures intricate relationships between body movements and foot positions, ultimately enhancing accuracy in classification and prediction tasks. Comprehensive experiments on real-world datasets demonstrate DGSA's robustness, generalization, and superiority in accuracy. Our approach achieves notable accuracy metrics: gait velocity (1.6 m/s), dynamic stability margin (5.6 cm), gait variability (2.4%), joint range of motion (56 degrees), dynamic balance index (0.4), minimum toe clearance (2.3 cm), foot progression angle (8.6 degrees), and dynamic joint stiffness (172). This study includes a comparative analysis of gait analysis approaches based on these key performance metrics, demonstrating DGSA's significant advancement in gait analysis methodology.