FORECASTING ISOKINETIC MUSCLE FATIGUE UTILIZING MECHANOMYOGRAPHY SENSORS

Background: A mechanomyography (MMG) sensor serves as an alternative tool to describe muscle fatigue. This study aims to establish a relationship between muscle contraction signals and torque during a fatiguing isokinetic muscle contraction. Our research is focused on developing ANN prediction models based on muscular contraction and torque. Methods: Muscle contraction and torque signals were simultaneously collected from healthy subjects, each performing 50 maximum effort repetitive isokinetic knee extensions. The gathered signals underwent filtering and normalization. Subsequently, the peak contraction values of both signals were categorized into four distinct muscle fatigue levels. A significant decrease in the mechanomyography sensor voltage and torque value over time was observed. A prediction model was developed using the Levenberg–Marquardt algorithm to anticipate various muscle fatigue stages, including fresh, reduced, pre-fatigue, and fatigued stages. The input parameters for model development comprised average muscle contraction and torque data, while binary-formed fatigue levels served as output parameters. Results: The optimal model was configured with hidden layers consisting of 10 neurons, a threshold of 0.5, and a k-fold value of 5. The developed prediction model demonstrates a noteworthy correlation between muscle torque fatigue and muscle contraction fatigue, with a sensitivity of 80.55%. Conclusions: This study underscores the predictability of muscle fatigue based on the direct measurement of muscle contraction changes.