Classifying Motor Preparation and Execution of the Left and the Right Lower Limb Using Brain Network Features

Abstract

Upper limb motor-related cortical activity (MRCA) has been widely developed and applied in the BCI technology, but the lower limb MRCA is difficult to classify or transfer to control command due to the close anatomical distance of the motor area in human cortex. In previous studies, the classification system based on the cortical functional connectivity features has been proven to be able to classify the lower limb motor imagery recorded from clinical-grade EEG system. The aim of this study is to use the brain network features to distinguish lower limb MRCA that can be potentially applied to reality in real time. The movement-related cortex areas of the lower limbs are very close, making it difficult to classify using spatial features. By using the Pearson correlation to calculate the connectivity strength between EEG signals as brain network features, and in combination with linear SVM for classification, three subjects can be trained for less than twenty minutes and achieve an average accuracy of 77.92%. In offline testing, the average accuracy of cross-validation also reached 75.31%. Our research proves that the classification system constructed by the cortical functional connectivity features can operate in a real environment and has promising classification accuracy. This shows the potential application of our system for lower limb rehabilitation neural feedback system or lower limb BCI-based robotic control device.