A fixed confidence rule did not show a clear site identification gain over matched random rejection within five SEEG sessions

We tested whether a fixed confidence rule can improve identification of a stimulated contact pair from a SEEG response. We used 1,923 public pulses from 185 sites in five patients in OpenNeuro ds004457 version 1.0.2. At each site, four pulses formed a waveform template. Two more pulses set a patient confidence threshold. The other 813 pulses formed the test set. The main rule used the gap between the two largest cosine similarities at 15 to 100 ms. It retained 45.8 percent of site-weighted test pulses. Accepted accuracy was 25.6 percent. Random rejection with the same accepted count at each site had expected accuracy of 25.0 percent. The mean gain was 0.7 percentage points. The descriptive 95 percent patient interval was -0.8 to 2.1 points. The exact two-sided patient sign-flip p value was 0.4375. The result is inconclusive for a small benefit. A second confidence score gave a larger apparent gain, but most of that change came from the accepted site mixture. The split was within the same stimulation blocks. It does not test live or separate-session use. The package contains three figures, all predictions, an exact matched-control formula, source data bins, code, hashes, and retrieval instructions. A full raw reconstruction exactly matched all 150 stored arrays.

stereoelectroencephalography · SEEG · electrical stimulation · site identification · confidence rejection · matched random control · intracranial EEG · waveform classification