Late delay frontal high gamma did not lower mean error on later trials in 11 patients with fixed channel models

We tested whether late delay frontal high gamma improved prediction of later object memory scores. We used public SEEG from OpenNeuro ds006136, release 1.0.1. The fixed test used 48 channels from 11 patients. Training used the first two thirds of source trial IDs. Testing used 551 distinct later patient trials. Each channel model used trial order, fixation power, and encoding power. Adding late delay power changed mean squared error from 0.35394 to 0.35623. Mean gain was -0.00230 score squared, with an approximate 95% interval of -0.00864 to 0.00404. The exact sign-flip p value was 0.58496 under independent, equally likely null signs. Six patients improved. All planned sensitivity intervals included zero. This result concerns fixed linear channel models and accepted trials from the same patients. It does not rule out memory effects or gains from other models. The package preserves code, trial features, predictions, source versions, hashes, and reproduction instructions.

working memory · stereoelectroencephalography · high gamma · trial prediction · ridge regression · OpenNeuro