An investigation on dimensionality reduction in the source-space-based hand trajectory decoding

In this work, the hand trajectory decoding was investigated in the source space. A couple of di-mensionality reduction techniques were utilized to re-duce the number of the source-space signals, namely, computing the mean, principle component analysis (PCA), locality preserving projection (LPP).
The decoding performances from the source-space approaches were compared to the sensor-space ap-proach. We found that every approach showed perfor-mance metrics in a similar range and only slight differ-ences across approaches could be observed. The source-space approach with PCA with 8 components exhibited higher performance metrics than other ap-proaches and slightly higher performance metrics than the sensor-space approach (improvement for correla-tion 0.01 to 0.09, SNR 0.05 to 0.1 dB). The results sug-gested that the source-space-based decoding is pos-sible, and it can achieve comparable performance to the sensor-space approach.