Abstract
In this paper, we propose an artificial intelligence (AI)-based sarcopenia diagnostic technique for stroke patients utilizing bio-signals from the neuromuscular system. Handgrip, skeletal muscle mass index, and gait speed are prerequisite components for sarcopenia diagnoses. However, measurement of these parameters is often challenging for most hemiplegic stroke patients. For these reasons, there is an imperative need to develop a sarcopenia diagnostic technique that requires minimal volitional participation but nevertheless still assesses the muscle changes related to sarcopenia. The proposed AI diagnostic technique collects motor unit responses from stroke patients in a resting state via stimulated muscle contraction signals (SMCSs) recorded from surface electromyography while applying electrical stimulation to the muscle. For this study, we extracted features from SMCS collected from stroke patients and trained our AI model for sarcopenia diagnosis. We validated the performance of the trained AI models for each gender against other diagnostic parameters. The accuracy of the AI sarcopenia model was 96%, and 95% for male and females, respectively. Through these results, we were able to provide preliminary proof that SMCS could be a potential surrogate biomarker to reflect sarcopenia in stroke patients.