Automatic estimation of upper limb angular displacement based on EMG or EEG signals

Abstract

Upper limb disabilities affect each year a considerable portion of the population, which leads us to the need to assist those who totally or partially lost mobility in the upper limb. Most recent approaches have reconstructed the trajectories from muscular and neural signals into upper limb kinematics employing machine learning. This thesis aimed to estimate the elbow joint angle based on electromyographic (EMG) and electroencephalography (EEG) signals using signal processing and machine learning techniques. Twenty-one subjects (ten females, eleven males) were told to perform synchronic flexion-extension movements while EMG and EEG signals along with elbow angle were recorded. Recorded signals, EMG and EEG, were used to estimate the elbow angle employing a Multilayer Perceptron and a Long Short-Term Memory neural network. Long Short-Term Memory neural network outperforms the Multilayer Perceptron in all cases. Long Short-Term Memory obtained the best result by training one network per subject (intra-subject). The lowest error was reached using the EMG signal, obtaining a root mean square error (RMSE) of 8.59 and a coefficient of determination (R2) of 0.95. Employing EEG signals produced an RMSE = 9.27 and R2 = 0.95. When both signals, EMG / EEG, were used, the results were RMSE = 9.53 and R2 = 0.95. For the intersubject data (i.e., one model for all women and one model for all men), we obtained the lowest RMSE values using the combination of EMG / EEG signals, for both women and men, RMSE = 10.96 and RMSE = 9.92, respectively. The methodology proposed proved suitable for estimating angular values of the elbow joint by obtaining an RMSE < 9 utilizing EMG signals and intra-subject data. The approach adopted is also feasible using EEG signals, as they resulted in good performance RMSE < 10 , considering intra-subject data. Considering inter-subject data, an RMSE < 11 was obtained using the EMG / EEG combination. A new methodology is proposed for estimating elbow angles based on EMG and EEG signals. Being useful for generating control signals for prostheses and/or exoskeletons designed to provide the support that people with motor disabilities require.