Delay SEEG power gave an uncertain memory prediction gain with fixed logistic models on separate trial blocks in 13 patients

We tested whether delay SEEG power improved prediction of complete object memory success. We used OpenNeuro ds006136 version 1.0.1. The source supplied 143 bipolar channels and 2,134 session trials from 13 patients. We matched channels by original trial ID. Each session had five separate test blocks. Training trials within five trial IDs of a test block were excluded. Fixed logistic models used fixation power, encoding power, and trial order. The added model also used early and late delay power. The mean patient Brier score fell from 0.1986 to 0.1965. The gain was 0.0022, with a 95% patient bootstrap interval from -0.0015 to 0.0064. The exact sign flip test gave p = 0.348. Six patients improved. A stronger fixed penalty gave a larger gain in a sensitivity test. The main evidence remains uncertain. It applies to the supplied processed trials and this model. It does not test new patients, future trials, or burst timing. The package preserves aligned features, all predictions, source hashes, independent checks, and a raw data retrieval recipe.

SEEG · working memory · object memory · Brier score · logistic regression · prediction · delay power