A Machine Learning Framework for Gait and EMG Analysis for Post-stroke Motor Dysfunctions Assessment

A stroke is a notable medical disorder characterized by the abrupt cessation of blood circulation to the brain, causing a deficiency of oxygen and nutrients to brain tissue, which leads to cellular damage or demise. The effects of stroke on individuals may range significantly, from modest impairments to profound disability. Stroke treatment entails extended therapy, often emphasizing gait rehabilitation. Gait rehabilitation programs focus on enhancing gait symmetry, velocity, and independence, since these qualities have a substantial impact on the likelihood of patients reintegrating into their pre-existing surroundings. Significantly, the assessment of muscle activation patterns and neuromuscular control using Electromyography (EMG) measurement during walking is of utmost importance for stroke patients. Machine learning methods have the capacity to provide useful insights on the patterns of walking and muscle activation, functional results, and strategies for rehabilitating stroke patients. The objective of this research is to use a Support Vector Machine classifier to categorize individuals with Stroke from healthy controls by analyzing gait and EMG measures, both alone and in combination. The best performances were obtained employing both gait and EMG features, reaching a test accuracy of 91.4%. The findings can foster the employment of ML approaches to help clinicians to diagnose stroke-dependent motor impairments.