Published: 27-11-2020| Version 1 | DOI: 10.17632/jhv73gj2d2.1
shixian liu


The data included stroke patients and normal subjects. When they performed motions of hand, the prefrontal hemodynamic data were recorded.


Steps to reproduce

Discrete cosine transform (DCT) was used for filtering. The low pass filter was used to remove physiological interference, and the high cut-off frequency was set to 0.1 Hz. The high pass filter was used to remove baseline drift, and the low cut-off frequency was set to 0.004 Hz. Concentration was calculated by Modified Beer-lambert Law (MBLL). Further, after paired-sample t-test, there was no significant difference between the left and right hands of normal subjects, so the two were compared as a group. And then activation rate of affected hand, activation rate of unaffected hand and activation rate of hand of normal subjects were compared with one-way ANOVAs. Bonferroni’s multiple comparisons test was used to compare the differences between any two of three groups. All statistical significance was set at p<0.05. Finally, time series relative change values, ΔHbO_2 and ΔHbT was divided into several blocks with the length of one second. Classification using SVM with RBF kernel was performed to distinguish 5 tasks. Two-thirds of the data was used for training, and the rest for prediction. Because the number of tasks is 5, the BCI needed to be designed as a multi-classifier. However, SVM is a two-classification algorithm, so that we adopted a one-versus-one strategy, which is a designed SVM between any two samples. Therefore, a N(N-1)/2 SVM should be designed for N class samples. When a predicted sample is classified, the category with the most votes is the class of the predicted sample.