Sarcopenia is the age related loss of skeletal muscle mass and strength, which increases frailty, fall risk, and healthcare costs. Conventional imaging based diagnostics are costly and impractical in routine clinical or community settings. We investigated whether noninvasive physiological signals, namely electrocardiography (ECG), electromyography (EMG), and center of pressure (COP), can serve as biomarkers for sarcopenia when analyzed with machine learning models. In this study, 80 Korean women aged 65 to 75 years were assessed for sarcopenia using the 2019 Asian Working Group for Sarcopenia criteria (skeletal muscle mass index, hand grip strength, Short Physical Performance Battery). ECG, EMG, and COP signals were recorded under standardized laboratory conditions. Features that differed between participants with and without sarcopenia were identified using the Mann Whitney U test and then used to train support vector machine, neural network, random forest, and gradient boosting classifiers. Stratified ten fold cross validation evaluated accuracy, F1 score, precision, recall, and the area under the receiver operating characteristic curve (AUC). Statistically significant differences ( p < 0.05) in physiological signal features were observed between sarcopenia and non-sarcopenia groups. ECG-based models consistently achieved the highest classification performance, with the SVM classifier reaching an accuracy of 93.2% and F1-score of 0.932. EMG- and COP-based models also demonstrated effective performance. Notably, for COP signals, models using only statistically significant features performed comparably or better than those utilizing the full feature set. Non-invasive physiological signals, particularly ECG-derived features, can effectively distinguish sarcopenia in elderly women in South Korea. Machine learning-based analysis offers a practical and scalable screening approach that complements existing diagnostic tools. This method has potential to enhance early detection efficiency and reduce diagnostic burden, especially in community and primary care settings. Not applicable.
Yoon et al. (Fri,) studied this question.