Beta-band activity was not significantly more robust than high-gamma activity to loss of the eight best SEEG channels

This study tested whether beta-band (13–30 Hz) movement information is distributed across more stereo-electroencephalography (SEEG) channels than high-gamma (70–150 Hz) information, making beta decoding more robust when the best channels are unavailable. Archived features from eight subjects in OpenNeuro ds004473 v1.0.2 were evaluated with shrinkage linear discriminant analysis and a virtual lesion that removed the k channels with the largest absolute paired t statistic in each training fold. The planned primary endpoint was the difference between bands in the AUC loss from k=0 to k=8. Mean loss was 0.114 for high gamma and 0.074 for beta; the mean difference was 0.040 (high gamma minus beta), with six of eight subjects showing a larger high-gamma loss. This difference was not significant (exact two-sided sign-flip p=0.21). At k=8, median AUCs were 0.713 for high gamma and 0.685 for beta; the band difference was not significant. Secondary analyses also did not show significant beta advantages in the fraction of significant channels, the number of shafts with significant channels, or along-shaft correlation. An HG+beta decoder lost less than HG alone in seven of eight subjects, but with eight channels removed its AUC remained below HG with all channels (median difference −0.039; p=0.047, uncorrected). The top eight channels overlapped by only zero to three channels per subject. In these eight subjects and this offline, single-session cued key-press task, beta information was not significantly more robust to loss of the best channels and did not compensate for the removed channels. The small sample, low power, and virtual-lesion design limit generalization to implants that miss motor cortex.

stereo-electroencephalography · SEEG · beta band · high gamma · virtual lesion · brain-computer interface · shrinkage LDA · channel robustness