A pre-registered test did not replicate pre-movement choice decoding in five new SEEG patients

A pre-registered replication tested whether left/right choice could be decoded from 70–150 Hz stereo-electroencephalography (SEEG) high-gamma activity in the 0.5 s before movement onset. Using the previously reported shrinkage linear discriminant analysis on five new patients, the mean pre-movement area under the ROC curve (AUC) was 0.553; four of five patient AUCs exceeded 0.5, but the one-sided exact sign-flip test did not pass (p = 0.156). One planned patient was not analyzed because of the run-time limit. In the pooled 12 patients, including seven from the earlier report, the mean AUC was 0.541 (11/12 above 0.5; p = 0.021). The pre-stimulus control was near the pooled pre-movement result (mean AUC 0.527; p = 0.063), and error-trial analyses did not establish that the decoder followed choice rather than stimulus. The evidence for a choice-specific signal is weak. This offline analysis used author-provided high-gamma derivatives, one time-ordered 60/40 split per patient, and five new patients; the test had low power, patient selection was by file size rather than randomization, and no channel localization was used. For a brain-computer interface, the measured pre-movement signal is at most a weak and unreliable left/right intent signal.

stereo-electroencephalography · high-gamma activity · choice decoding · pre-registered replication · brain-computer interface · shrinkage linear discriminant analysis · OpenNeuro ds006253