Predicting force loss during dynamic fatiguing exercises from non-linear mapping of features of the surface electromyogram

This study proposes a method for estimating force loss during fatiguing maximal isokinetic knee extension contractions using a set of features from surface EMG signals recorded from multiple locations over the quadriceps muscle. Nine healthy participants performed fatiguing tests which consisted of 50 and 75 isokinetic leg extensions at a speed of 30°/s and 80°/s in two experimental sessions on different days. The set of data recorded from one of the experimental sessions (at both velocities) was used to train a multi-layer perceptron neural network to associate force loss (direct measure of fatigue) to EMG features. The data from the other session (obtained from the tests at both velocities) were used for testing the neural network performance. The proposed method was compared with a previous approach for the assessment of fatigue (Mapping Index, MI) using a signal to noise metrics computed on the estimated trend of fatigue. The signal to noise ratio obtained with the proposed approach was greater (8.83 ± 1.07) than that obtained with the MI (5.67 ± 1.17) (P < 0.05) when the subjects were analyzed individually and when the network was trained over the entire subject group (8.07 vs. 4.42). In conclusion, the proposed approach allows estimation of force loss during maximal dynamic knee extensions from surface EMG signals with greater accuracy than previous methods.