Maps of 30–80 minus 1–30 Hz correlation strength reduced mean error after shaft name and distance controls in separate SEEG files from 36 epilepsy patients

We tested whether frequency band contrasts predict measured SEEG link maps after fixed geometry controls. Public OpenNeuro ds004100 version 1.1.3 supplied two interictal files from each of 36 epilepsy patients. The contrast was absolute correlation at 30–80 Hz minus absolute correlation at 1–30 Hz. Each link connected bipolar signals from different named shafts. We removed shaft-pair means and a quadratic trend in log distance from source coordinates. A weight from two training blocks scaled the training map. The target was the residual map in the separate file. Mean squared error fell from 0.003622301 for zero residual to 0.002368560 for the weighted map. The reduction was 34.61 percent. Its descriptive 95 percent patient bootstrap interval was 22.26 to 43.96 percent. Gains were positive in 34 of 36 patients. Population use of the interval requires independent and exchangeable patients. All nine planned checks had positive mean gains. A fixed half weight had slightly lower mean error than the fitted weight. All 72 covariance inputs were rebuilt from pinned raw ranges and matched exactly. Independent matrix checks and complete analysis repeats passed. The package preserves code, inputs, fitted maps, source hashes, results, and retrieval instructions. This is a new analysis of an existing cohort. File timing and possible partial overlap remain unknown. Repeatable artifacts can also produce band contrasts. The result does not establish neural connections or clinical benefit.

stereo-electroencephalography · intracranial EEG · band contrast · correlation · network measurement · patient bootstrap