چكيده به لاتين
In the last few decades advancement in Neuroscience and Engineering has provided brain computer interface (BCI) as a tool for rehabilitation research. One of the most important types of BCIs is movement control-based BCIs that, by estimating the movement parameters from brain signals, can control external devices such as the artificial limb. In the past, the decoding of movement from invasive brain signals was well done and had satisfactory results. But in recent years, because of the benefits of non-invasive brain signals, researchers have turned to non-invasive recording methods such as Electroencephalogram (EEG). In this study trajectory of movement was decoded, using EEG signals simultaneously measuring the movement of the hand, in a continuous curve movement task in the horizontal plane along two axes X and Y. The feature used is the signal modulation for various frequency bands, channels, and time lags. In the following, a subset of optimal features is selected by the mutual information, and with the measured data is given to the repressor to learn the model. The mean correlation coefficient (± SD) for decoding in two axes X and Y of the recorded signal from 10 healthy subjects for PLS, KRR and MLR regressors was 0.33 ± 0.06 and 0.57 ± 0.19, 0.32±0.07 and 0.55±0.07, 0.32±0.06 and 0.55±0.19, respectively. PLS was performed better than others but not significantly. During studies performed on different frequency bands, gamma band (above 30 Hz) had the most effect on feature selection. Finally, the velocity decoding was also calculated and the mean correlation coefficient was obtained 0.18 and 0.28 on both axes X and Y, respectively. According to the results, there is a possibility of decoding the movement parameters from the EEG signal, and using the method used in this study can be achieved relatively acceptable performance compared to the past.
