Gait Analysis-Based Identification of Neurodegenerative Diseases Using Machine Learning Techniques

The current study introduces an improved framework for assessing neurodegenerative diseases by evaluating gait data using machine learning techniques and designing a wireless Velostat-based insole along with a gait phase detection algorithm. The machine learning model was trained using vertical ground reaction force datasets available on PhysioNet. The first dataset includes the data of 73 healthy individuals and 93 individuals with Parkinson's disease. Another dataset used in the present study consisted of 16 healthy subjects and 15, 20, and 13 subjects with Parkinson's, Huntington's, and amyotrophic lateral sclerosis, respectively. Upon testing, the classification accuracy achieved was 99.08% and 99.95% on both datasets, using the KNN classifier. Finally, this trained model was used to identify neurodegenerative diseases, where the data was gathered from a Velostat insole system created in-house. The current method has been shown to demarcate gait events reliably. Authors also intend to employ it in clinical research by obtaining vertical ground response force data from patients with neurodegenerative diseases. In conjunction with the gait recognition algorithm, the developed insole system reliably marks the following gait events: (i) vertical ground response force; and (ii) Spatio-temporal gait characteristics.