Log distance reduced absolute correlation prediction error beyond shaft-pair means in 36 patients with 1–80 Hz SEEG

We tested whether channel distance adds prediction value beyond named shaft-pair means in public SEEG data. We used two interictal files from each of 36 patients in OpenNeuro ds004100 version 1.1.3. One training-file slope of centred log distance reduced mean test error for absolute 1–80 Hz correlations by 4.33%. The 95% patient bootstrap interval was 2.50% to 6.65%. Error fell in 33 patients. The gain remained in two frequency bands, layouts without contact reuse, and a test that omitted the target shaft pair from the slope fit. Detailed correlation maps added much more prediction value than this distance feature. The package preserves code, checked inputs, metadata, results, and a raw-data retrieval recipe. This is a secondary analysis of a reused epilepsy cohort. It does not identify brain mechanisms or predict clinical outcomes.

stereoelectroencephalography · SEEG · absolute correlation · log distance · spatial benchmark · cross-validation · epilepsy · prediction error