Detection of Affected Anatomical Systems in Patients with Gait Disorders Using Artificial Intelligence

Abstract

Gait disorders are a multidisciplinary clinical problem resulting from the involvement of a wide anatomical spectrum ranging from the central and peripheral nervous systems to extrapyramidal structures, the musculoskeletal system, and joint-bone pathologies. This study aims to automatically classify the affected anatomical system by analyzing high-resolution clinical gait videos using a 3D Convolutional Neural Network (3D-CNN) architecture. The video-based dataset collected at the Neurology Clinic of Ondokuz Mayıs University was labeled under four categories: central nervous system, peripheral nervous system, extrapyramidal system, and musculoskeletal-joint pathologies. The videos were processed through an OpenCV-based pre-processing pipeline, spatial and temporal standardization was applied, and the model was optimized with an 80% training – 20% test split. The 3D-CNN model achieved an accuracy level of 96.20% after 50 epochs, demonstrating high performance in terms of clinical generalization. The findings demonstrate that video-based deep learning approaches can enhance early diagnosis, risk reduction, and personalized treatment planning in decision support mechanisms related to neurological and musculoskeletal disorders. Unlike the mostly single-disease focused models in the literature, this study offers a comprehensive clinical classification with multiple class distinctions and provides a strategic foundation for the integration of AI-based gait analytics into real-world applications.