Median templates raised four-site SEEG accuracy from 71.9% to 75.6% in eight patients with five fit pulses per site and fixed four-pair scales

We tested whether the median of five fit pulses improves a simple SEEG reader. We used OpenNeuro ds005448 release 1.0.0 and a fixed sample from an earlier sci publication. Each of eight patients had four stimulation sites and five separate test pulses per site. Both readers used the same four recording pairs, feature scales, and 20 to 100 ms window. Only the template center changed. The mean reader identified 115 of 160 test pulses (71.9%). The median reader identified 121 (75.6%). Four patients improved and four stayed the same. The exact patient sign-flip test gave p = 0.125. The patient bootstrap interval was +0.625 to +7.5 percentage points. The median did not improve the average worst accuracy after one fit pulse was removed. It also did not reduce scaled feature error. A secondary reader that permits recording-pair sign changes lost 2.5 points. This reused, offline sample does not establish a benefit for new patients or a live BCI. The package contains figures, every decision, compact data, source receipts, code, and exact repeat checks.

SEEG · stimulation site identification · median template · intracranial EEG · signal processing · template classification · reproducibility