Predicting stroke, neurological and movement disorders using single and dual-task gait in Korean older population

Single and motor or cognitive dual-gait analysis is often used in clinical settings to evaluate older adults affected by neurological and movement disorders or with a stroke history. Gait features are frequently investigated using Machine Learning (ML) with significant results that can help clinicians in diagnosis and rehabilitation. The present study aims to classify patients with stroke, neurological and movement disorders using ML to analyze gait characteristics and to understand the importance of the single and dual-task features among Korean older adults.A cohort of 122 non-hospitalized Korean older adult participated in a single and a cognitive dual-task gait performance analysis. The extracted temporal and spatial features, together with clinical data, were used as input for the binary classification using tree-based ML algorithms. A repeated-stratified 10-fold cross-validation was performed to better evaluate multiple classification metrics with a final feature importance analysis.The best accuracy - maximum >90 % - for gait and neurological disorders classification was obtained with Random Forest. In the stroke classification a 91.7 % of maximum accuracy was reached, with a significant recall of 92 %. The feature importance analysis showed a substantial balance between single and dual-task, while clinical data did not show elevated importance. The current findings indicate that a cognitive dual-task gait performance is highly recommendable together with a single-task in the analysis of older population, particularly for patients with a history of stroke. The results could be useful to medical professionals in treating and diagnosing motor and neurological disorders, and to improve rehabilitation strategies for stroke patients. Furthermore, the results confirm the proficiency of the tree-based ML algorithms in biomedical data analysis. Finally, in the future, this research could be replicated with a non-Asian population dataset to deepen the understanding of gait differences between Asian-Korean population and other ethnicities.