Adding current baseline power reduced mean prediction error for later response power by 0.075 percent in eight SEEG patients with the specified ridge rule

We tested if current baseline SEEG power improves response power prediction beyond task conditions and recent trial history. We used OpenNeuro ds004473 version 1.0.1 with eight patients and 2,030 aligned trials. Models used 973 earlier training trials and 1,017 later test trials. Adding the current baseline map gave a 0.075072 percent mean error reduction. Seven patients had lower error. The 95 percent patient bootstrap interval was 0.009768 to 0.155461 percent. The exact two-sided patient sign-flip test gave p = 0.0546875. This test did not reject zero gain at a 0.05 threshold under the symmetric null assumption. All primary map models selected penalty 100 from the specified grid. A planned fixed penalty of 10 increased mean error by 0.577818 percent. The gain was small and depended on the fitting rule. These data do not establish a useful gain from the added baseline map. All 4,060 selected source epochs were rebuilt from fixed public EDF object versions. Every feature and range hash matched. Independent reader and numeric audits passed. The package preserves inputs, code, fitted models, predictions, all planned checks, and source retrieval instructions. Response times were known retrospectively. The test does not establish clinical benefit or prediction in new patients.

stereoelectroencephalography · SEEG · spectral power · ridge regression · response prediction · baseline power