Finger rehabilitation training and evaluation system based on EEG signals, machine learning and Fugl-Meyer scale

For patients with digital hemiplegia caused by stroke, equipment such as mechanical exoskeleton devices are currently used to help recovery, but they are limited to many shortcomings of this method. This paper proposes to collect electroencephalogram(EEG) information through brain-computer interface (BCI) equipment. Combined with the research status of BCI system at home and abroad, we established the research idea of BCI system based on motor imagery by preprocessing the obtained information. According to the biological characteristics of human fingers and the needs of finger rehabilitation, a finger rehabilitation system was designed to assist hemiplegic patients in rehabilitation training. In the experiment, the EEG signals of several subjects in the two states of imagining finger movement and rest were collected through the EEG cap, and an appropriate feature extraction method was selected. Machine learning like logistic regression, random forest and deep learning were used for various classifications. The EEG feature vectors extracted by different subjects were classified and cross-validated. At the same time, the popular Fugl-Meyer scale was selected to quantitatively assess the patient’s limb function. The experimental results show that the comprehensive classification effect of deep learning is the best. The system performed well for the patient’s recovery process.