In 15 SEEG patients, 8 channels did not meet non-inferiority for movement decoding; 32 channels did

We retested whether eight training-selected SEEG channels decoded movement versus rest as well as all channels in 15 patients from OpenNeuro ds006253. We used archived features, new analysis code, one log10 window mean of 70–150 Hz high-gamma activity per channel, and a time-ordered split (first 60% of trials for training, last 40% for test). The plan was written and hashed before analysis. With all channels, median test AUC was 0.836. For eight selected channels, the mean difference from all channels was −0.025 AUC (bootstrap 95% CI −0.039 to −0.010); 8 of 15 patients were within the 0.02 margin. Non-inferiority was not shown (exact sign-flip p = 0.73). The seven patients in an earlier SCI result had a mean difference of −0.024, and the eight other patients had −0.025. Five-fold cross-validation gave a similar result (−0.025, p = 0.83). Thirty-two selected channels met non-inferiority (mean −0.003, p = 0.003). Eight selected channels outperformed eight random channels by 0.158 AUC across all 15 patients (p < 0.001). This retest reproduces the earlier result in seven overlapping patients, with a larger sample and a different pipeline. It is an offline analysis of one task, and the movement window followed the stimulus and decision. For this task, plan for about 32 recording contacts or validate the channel count on the target task and patient.

stereo-electroencephalography · SEEG · movement decoding · brain-computer interface · channel selection · high-gamma activity · non-inferiority · AUC · shrinkage LDA · OpenNeuro