Absolute cosine reduced mean four-site accuracy in eight STReEF SEEG patients after recorded contact order reversal

We tested whether absolute cosine improves site identification after recorded stimulation contact order reversal. We used public STReEF SEEG data from eight patients with epilepsy. Each patient supplied four sites, four recording pairs, and five fit pulses per site. Testing used 160 separate pulses in total. The feature window was 20 to 100 ms. Signed cosine gave 100 correct decisions; absolute cosine gave 91. Mean accuracy decreased from 62.5% to 56.9%. The exact paired patient test gave p = 0.125. Absolute scoring gained seven correct answers and lost sixteen. Artificial global sign changes left all absolute decisions fixed. Recorded order reversal did not produce a uniform global response sign change. Only 40 test pulses occurred after all fitting pulses for their patient. The sample and earlier signed results were known. The result does not establish a population effect or live BCI performance. We preserve three figures, code, fitted readers, pulse decisions, source hashes, and reproduction instructions. Fresh retrieval matched all 320 raw epochs. A new Python environment repeated the analysis and matched the preserved inputs and results.

stereoelectroencephalography · electrical brain stimulation · cosine similarity · site classification · brain-computer interfaces · signal processing