Enhancing Classification of Grasping Tasks Using Hybrid EEG-sEMG Features

Systems based on multimodal Human-Machine Interface (HMI) propose a significant advance in rehabilitation engineering. This advance is due to their capacity to acquire information from different sources, allowing greater patient adaptability to the system in daily activities, and a higher accuracy rate in decoding the motor intention. Nowadays, there is the challenge of implementing better techniques to increase the classification rate of HMIs. This paper describes the implementation of two techniques based on a hybrid sEMG and EEG method to improve the classification of three types of grasp (spherical, cylindrical, and lateral). The proposed methods were evaluated on a public database of 25 subjects by using 11 EEG channels and 6 EMG channels, where the signals were segmented into five different time windows. The results show maximum accuracies above 70% for detecting grasping versus resting movements and accuracy above 54% for differentiation between each grasping movement. This performance improvement is obtained in time window between 1 and 2.5 s. The results allow us to conclude that the use of multimodal information allows an improvement in the identification of grasping tasks, and identifies a suitable time window that can be used to improve the performance of a multimodal system in real-time.