Advancing Quadriceps Muscle Monitoring: Wearable A-mode Ultrasound and Machine Learning Classification for Accurate Estimation of Muscle States

Accurately estimating muscle states, including fatigue and contraction, holds significant potential in the fields of rehabilitation and muscle-related disorder identification. However, conventional invasive methods such as intramuscular electromyography (EMG) entail risks and discomfort. This study pioneers the integration of A-mode ultrasound (US) signals from a wearable sparse array with machine learning (ML) classification, advancing quadriceps muscle state estimation. A novel wearable transducer, comprising a 16-element array with a central frequency of 10 MHz, enables the capture of A-mode US signals on curved skin surfaces. Three able-bodied participants engaged in two experimental sets: voluntary knee extension for contraction prediction, and a fatiguing protocol utilizing functional electrical stimulation (FES). Employing US feature extraction, followed by supervised ML classification, resulted in an exceptional average accuracy of 93.66 % in contraction classification and achieved 90.1 % in fatigue classification.