A SEEG reader kept an after-pulse site accuracy gain after removal of linear trend use in five offline patient sessions

We tested whether a SEEG site reader needs linear voltage trends. We used public OpenNeuro ds004457 version 1.0.2 and a fixed benchmark from five patient sessions. Four early pulses per site fitted each reader. Four later, distinct pulses tested it. There were 183 sites and 732 test pulses. The distance rule removed constant offsets and linear trend use at each recording pair. It used actual mean sample times in each voltage bin. Mean accuracy was 38.51% from 20 to 100 ms after a pulse and 3.48% in an equal-length window before it. The mean gain was 35.03 percentage points. Each session had a positive gain. The descriptive 95% bootstrap interval was 26.76 to 44.86 points. The exact two-sided sign-flip p value was 0.0625 under the stated null assumptions. The ordinary reader scored 40.56% after the pulse. Thus, trend removal had a mean accuracy cost of 2.04 points. Added digital affine voltages left every tested trend-removed decision fixed. We checked all 110 source file hashes, rebuilt all inputs from raw MEF data, and repeated all numerical outputs. The package preserves data bins, code, fitted models, scores, figures, proof, and retrieval instructions. Fit and test blocks interleave across sites. Only 36 of 732 tests occurred after the full library was fitted. The result does not establish a neural source, prospective BCI performance, or useful brain writing.

SEEG · intracranial EEG · electrical stimulation · site classification · linear trend removal · covariance distance · brain-computer interface