Data-driven estimation of EMG muscular activity and fatigue through infrared thermal imaging

Purpose: Superficial electromyography (sEMG) is the recording, from the surface of the body, of the electrical signal associated to muscle activation. Usually, sEMG is assessed through electrodes with electrolytic gel often causing skin irritation. To overcome this issue, capacitive contactless electrodes have been developed. However, contactless EMG sensors are still quite sensitive to motion artifacts and could be not comfortable for long monitoring. In this study, a non-invasive contactless method to assess muscular activity through infrared thermal imaging (IRI) is presented. Methods: 10 healthy participants (age: 21.8 ± 2.9 years) were enrolled in the study. The participants underwent to 5 series of bodyweight squat exercise until exhaustion separated by 1 min of rest. The vastus medialis activity was assessed through EMG system Encephalan Mini AP-10. Concurrently, the temperature of the same muscle was measured through thermal camera FLIR A655. Regarding the EMG, the Average Rectified Value (ARV) and the median frequency of the Power Spectral Density (MDF) were evaluated for each series. Specifically, ARV is indicative of muscular activity and MDF of the muscular fatigue. Concerning the IRI, the average and the standard deviation of the temperature in a temporal window of 10 s after each series, and the thermal spatial gradient of the considered region were computed. Several Machine Learning regressors were tested employing the IRI features as input and, separately, the ARV and MDF as output. The data were normalized (z-score) and the leave-one-subject-out cross validation was used to test the generalization performance of the models. Results: Concerning the ARV, the Gaussian Process Regression delivered the best performance, with a correlation coefficient r = 0.75 (p\ 0.001) and root mean square error (RMSE) of 0.02 mV. Regarding the MDF, the Support Vector Machine with a radial basis function kernel allowed to obtain the best regression (r = 0.66, p \0.001; RMSE=0.67 Hz). Conclusion: The proposed method estimated the EMG parameters indicative of muscular activity and fatigue. These results indicate that the muscular activity influences skin temperature, suggesting a modification of the superficial blood circulation linked to the muscular need of oxygen during exercising. These results could pave the way to the employment of contactless methods to monitor the muscular activity and evaluate fatigue in a non-invasive and comfortable manner in sports and clinical applications.