Item high gamma gave no clear recall prediction gain beyond list averages in 18 public SEEG patients

We tested study item high gamma beyond list average power in 18 public SEEG patients. The source was OpenNeuro ds005059 version 1.0.6. This exploratory secondary analysis reused a prior AutoSci cohort and test lists. Each channel feature was split into a six-item list average and an item deviation. The deviation is unchanged by an additive channel-list log-power offset. Earlier lists selected neural penalties. Later lists tested the models. There were 1,404 training trials and 930 test trials. The patient mean Brier gain from item deviations was -0.001982. The 95 percent patient bootstrap interval was [-0.006229, 0.000534]. The exact two-sided sign-flip probability was 0.484375. Eleven models omitted the item feature. This test did not detect a clear added gain. It does not establish equal performance or absent neural information. The interval holds fitted models and recorded test lists fixed. The sign-flip test requires independent patients and symmetric null contrasts. The package preserves code, compact features, predictions, all model choices, source versions, hashes, checks, and reproduction commands.

stereoelectroencephalography · high gamma power · memory · recall prediction · list average · item deviation · Brier score · exploratory analysis