Publication:

Development of an Interface for Rehabilitation Based on the EMG Signal for the Control of the Ankle Exoskeleton T-FLEX

(Eng) Stroke is the second leading cause of death and third of disability, and 75% of individuals who sustain a stroke each year experience limitations in mobility-related to walking. Strategies involving robotic devices, such as exoskeletons and orthoses, have been considered to improve stroke rehabilitation. Some of them have included the implementation of Electromyography (EMG) signals either for muscle activation analysis or movement intention detection. The latter has been involved in the activation process of robotic devices to handle the device’s assistance by the subject’s intention to perform a specific movement. This would allow the subject to get involved in his/her therapy. Hence, this project introduces an EMG interface for the control of the ankle exoskeleton T-FLEX. Some studies where EMG signals have been included in control and therapy processes were reviewed, and algorithms with different threshold methods calculation were analyzed. Considering the information from those studies, a threshold-based algorithm for movement intention detection was developed. The algorithm consisted in two main stages, the threshold calculation and the movement intention detection. The first stage consisted on the threshold establishment through statistical features extraction (MEAN, standard deviation (STD), variance (VAR), MEAN + 3*STD and Root Mean Square value (RMS)) from the EMG signal. The second consisted of comparing the signal with the reference value (threshold). To test the algorithm, two sessions were planned. In the first session, ten healthy subjects participated and their EMG signal was acquired from the Tibialis Anterior muscle through a Myoware muscle sensor. Additionally, an Inertial Measurement Unit (IMU) sensor was placed on each participant’s foot tip to acquire the angular velocity when the ankle’s dorsiflexion was performed. The output signals from both sensors were recorded and the processing with the algorithm was done offline. The second session was carried out with the ankle exoskeleton TFLEX and a Serious Game, implementing the algorithm in real-time with a statistical feature selected from the first session as the threshold. The detection from the EMG algorithm was evaluated. The algorithm that T-FLEX already had for the movement intention detection with the IMU sensor also was evaluated. The results from the first session showed that the MEAN feature worked for the threshold establishment with the IMU sensor, and for the EMG sensor was the (VAR), presenting and error of less than 10% in the amount of False Positive (FP) and False Negative (FN) values. With this, the second session was carried out, showing that there was more precision handling the game using the IMU sensor than the EMG sensor. With the EMG sensor the maximum precision achieved was 89,7% and with the IMU sensor was 94.1%.