Adding 1 to 30 Hz maps to a fixed ridge model reduced 30 to 80 Hz SEEG residual error by 3.8 percent from run 1 to run 2 in 36 epilepsy patients

We tested whether a low-band SEEG map adds prediction after a high-band map was used. We used public data from 36 epilepsy patients in OpenNeuro ds004100 version 1.1.3. Each map used correlations between bipolar channels on different electrode shafts. We removed row and column means from each shaft pair map. We fit a fixed ridge model with two blocks from run 1 and tested it on run 2. Adding the 1 to 30 Hz map reduced mean 30 to 80 Hz residual error by 3.83 percent. Error fell in 33 of 36 patients. The mean gain was 2.03e-5 in squared correlation units. Its 95 percent patient bootstrap interval was 1.03e-5 to 3.51e-5. Intervals included zero when file order was reversed or shared contacts were removed. The interval also included zero when the low band was the target. A check added after the fixed test gave 2.96 percent less error than duplicated high-band predictors. All 72 covariance inputs were rebuilt from pinned public EDF ranges with exact array agreement. The package includes inputs, code, model predictions, scores, source pins, and repeat instructions. The result concerns recorded residual maps in this cohort. It does not show a neural cause or clinical benefit.

SEEG · stereoelectroencephalography · ridge regression · correlation maps · frequency bands · epilepsy