Mean SEEG key decoding across task conditions exceeded chance in eight patients under conditional label exchange

We tested whether a fixed SEEG decoder could identify the response key across recorded fast and slow task conditions. We used OpenNeuro ds004473 version 1.0.1, eight patients, and 2,030 trials. Test trials formed five time groups. Training trials had a 30 second gap from each test interval. The two training conditions had equal trial counts. Mean balanced accuracy across conditions was 54.97% (conditional shuffle p = 0.0005). This test assumes label exchange within each time and condition group. One patient passed correction across eight patient tests. An exploratory rotation test gave the same group p value. Fresh public data gave identical feature arrays. The package contains code, features, source records, predictions, label draws, and checks. The result does not establish movement speed, transfer between patients, or clinical use.

stereoelectroencephalography · response key decoding · task-condition transfer · balanced accuracy · conditional permutation test · intracranial EEG