Sarcopenia Risk Levels Recognition by Gait Kinetics Using Deep Learning

Sarcopenia is an age-related disease that reduces muscle mass and harms the mobility of the elderly. Early diagnosis and risk assessments can contribute to the intervention of the elderly with a risk of sarcopenia. Currently, most gait analyses focus merely on gait parameters, and the knowledge of how to assess the sarcopenia risk conditions from gait is still insufficient. In this study, a convolutional neural network (CNN) based approach was proposed to predict three risk levels of sarcopenia. Forty elderly subjects with the risk of sarcopenia participated in the experiment and the kinetics of their gaits were collected as input to the CNN model. Physical function assessments (Short Physical Performance Battery test, five times sit-to-stand, hand grip strength, and gait speed) on the risk levels of sarcopenia were used as labels. Gait results from an example elderly subject with severe risk of sarcopenia demonstrated an overload hip and ankle flexion moment during the double support and single support phase. Model training results showed that the proposed classifier could achieve an accuracy of 97.99% and its prediction confusion matrix had an accuracy higher than 94.5% along the diagonal line. These results indicated the potential of decoding sarcopenia risk status from elderly individuals’ gait and thereby facilitating customized intervention strategies.