Phase removal lowered mean 4–30 Hz SEEG site skill gain for nearest mean readers in eight reused records at 100–500 ms

We tested which SEEG data a stimulation response reader must retain. We used public OpenNeuro ds004696 version 1.0.1 and exact coefficient arrays from an earlier sci study. The eight patient records supplied 97 tasks and 430 distinct later pulses. Each task fixed a remote recording pair and current. We trained on five early pulses per target and tested five later pulses in the same target blocks. For nearest mean readers at 4 to 30 Hz and 100 to 500 ms, mean after-pulse accuracy was 34.6% with amplitude and phase, and 29.1% with amplitude alone. We scaled accuracy for the number of targets and subtracted a before-pulse control. Removal of phase lowered mean skill gain by 6.1 points. The descriptive patient bootstrap interval was 2.6 to 10.1 points. All eight patient differences were positive. A tuned ridge reader gave a smaller difference of 3.6 points, with an interval from -4.1 to 11.2. Thus, the phase result depends on the reader. The test does not show a benefit in a new session or prove a neural cause. The package contains coefficients, model states, predictions, figures, source hashes, a retrieval recipe, and code. A fresh source check for one patient and a full numerical replay passed.

stereoelectroencephalography · Fourier phase · electrical stimulation · site classification · brain-computer interface · phase removal · exploratory study