GaitMed: A Medical Gait Dataset and Benchmark for Musculoskeletal Disease Classification

Abstract

Gait analysis has become a powerful tool for understanding, diagnosing, and monitoring musculoskeletal disorders. However, existing gait datasets primarily focus on human identification tasks, with limited applications in the medical diagnostic domain. Even when gait datasets are available for medical diagnosis, they often target a single condition, lacking the diversity needed to develop generalized gait analysis models. In this paper, we build a medical gait dataset (GaitMed), a comprehensive multi-view gait dataset dedicated to the analysis of musculoskeletal disorders. GaitMed contains gait data from 62 patients with various common musculoskeletal conditions, captured from multiple viewpoints and represented in different modalities, including silhouettes, skeletons, and skeleton heatmaps. To fully utilize the rich information in GaitMed, we propose a model for musculoskeletal disease classification using gait recognition, named MSGait. This model is a multimodal fusion two-branch gait recognition model based on the attention mechanism, which can integrate features from different modalities and viewpoints efficiently. Extensive experiments demonstrate that MSGait achieves competitive performance (Rank-1: 84.47%, Rank-3: 97.20%) comparable to state-of-the-art methods, with statistically significant improvements over multimodal baselines validating effective dual-branch fusion architecture. Furthermore, we showcase the potential of GaitMed in advancing data-driven approaches for musculoskeletal disorder analysis. Our dataset and model provide valuable resources and unique insights for medical research and clinical applications of musculoskeletal disorders, facilitating future research in medical gait analysis.