NeuroGait: Identifying Cognitive-Motor Impairment Through Secondary Functions in Walking Behavior with AI Gait Analysis

Overview

Current diagnostic methods can only detect neurodegenerative conditions after symptoms emerge, when significant neurological damage has already occurred. This project investigates early signs of cognitive-motor impairment through secondary function gait analysis using Parkinson’s disease as a model condition. After developing a comprehensive insole-based system, gait metrics were collected from clinically diagnosed patients and matched controls under both primary and secondary function conditions. The effects of secondary functions were evaluated across four conditions: (i) uninhibited walking, (ii) walking with simple auditory task, (iii) walking with complex auditory task, and (iv) walking with arithmetic task. Using clinically validated data, multiple machine learning models were then developed for severity classification. The classification models yielded accuracies of 87%, up to 97% with data augmentation using synthetic minority oversampling technique. In parallel, optimization between model complexity and accuracy also yielded results of 94% with the use of only 13 gait features. Subsequently, several novel biomarkers of cognitive-motor impairment, such as forefoot delay and forefoot impulse fraction, were identified through analysis of variance and Shapley value interpretation. These findings were validated by the additional identification of existing biomarkers, such as center-of-pressure metrics, consistent with previous studies. Monitoring these key biomarkers can support clinical diagnosis of neurodegenerative diseases and enable quantitative evaluation of treatment effectiveness. Ultimately, this AI-assisted end-to-end gait analysis system provides a generalizable foundation for scalable, non-invasive tools to characterize the development and progression of cognitive-motor impairment in neurodegenerative diseases.