A fitted model of independent contacts increased mean later beta correlation error for shared-contact SEEG pairs in one-second windows from 21 patients

We tested beta correlation forecasts in 35 public SEEG records from 21 patients in OpenNeuro ds006519 version 1.0.0. For 980 pair occurrences with a shared contact, we fitted a model of independent contacts on an early one-second window. We tested a separate later one-second window before stimulation. Mean squared error was 0.23189 for the model and 0.16315 for zero correlation. The increase was 0.06874, with a descriptive 95 percent patient bootstrap interval from 0.01294 to 0.12388. The ratio of mean errors showed a 42.1 percent increase. The exact patient sign-flip test gave p = 0.02906 under patient independence and null symmetry. One-half of the early measured correlation gave error 0.08054. Signs were opposite to the model in both windows for 29.1 percent of pairs after patient averaging. The package preserves source hashes, signal segments, fitted models, all scores, two figures, code, and repeat instructions. All source hashes, direct native sample reconstruction, and independent score checks passed. This reused epilepsy cohort and offline test do not establish neural connections or interface benefit.

stereo-EEG · beta correlation · bipolar channels · contact model · signal forecasting · brain-computer interfaces