Early SEEG slopes predict later shape added by power averaging in 24 people after normalization at 20–100 Hz with 55–65 Hz removed

We tested how power averaging changes SEEG spectrum shape. We used 34 valid records from 24 people in OpenNeuro ds005624, version 1.0.0. We divided each pair spectrum by its geometric mean power in the test band. We then compared arithmetic and geometric power means across pairs. Straight log spectrum fits from the first minute predicted their centered difference in the last minute. The prediction reduced mean squared error by 48.2% against a zero prediction. The 95% person bootstrap interval was 27.0% to 89.8%. All 24 people had lower error. The added quadratic coefficient was positive in all 34 records. Two post-plan controls showed the scope of this result. Individual slopes gave 14.0% lower group error than a mean slope prediction from the other people. Carrying the observed early shape forward gave lower group error than the slope model. Original power levels gave gains in only 15 people and often gave negative added curvature. The package preserves spectra, source pins, code, checks, and results. This test shows a processing effect. It does not identify a neural mechanism.

stereoelectroencephalography · power spectrum · arithmetic averaging · geometric averaging · spectrum shape · frequency slope · signal processing