Figure 2
Comparative analysis of decoding muscle activity from EEG using BI-SNN and GLM approaches. (A) Normalised cross-correlation coefficients between the actual and predicted muscle activity from anterior deltoid (AD), brachoradial (B), flexor digitorum (FD), common extensor digitorum (CED) and first dorsal interosseous (FDI) muscles by BI-SNN and GLM methods. (B) Statistical distribution of the correlation coefficients by BI-SNN (blue) and GLM (black). (C) Calculation of the normalised cross-correlation coefficients between EMG activity of AD muscle and SPANAD population activity. Top: rectified (green) and smoothed (black) EMG signal from the AD muscle of participant 2, and the average convoluted spike sequence generated by the SPANAD (blue) population (The alpha kernel was used for spike convolution—refer Eq. (8)), middle: convoluted spike sequences emitted by spiking neurons in SPANAD population, bottom: normalised cross-correlation coefficients between the smoothed rectified EMG signal and the average SPANAD convoluted population activity using cross-correlation lag of 200 ms. (D) Comparison of the mean cross-correlation coefficients by the BI-SNN and GLM approaches (statistically significant differences are highlighted in red). (E) Band-specific cross-correlation coefficients of participant 2. (F) Actual muscle activity of B, FD, CED, FDI muscles and the response of corresponding SPAN population (SPANB, SPANFD, SPANCED, SPANFDI). (G) Population activity of SPANmovement-onset and the accumulated amplitude of the five EMG sensors.
