Fitted SEEG slopes account for lower prediction error across labeled sessions in seven participants at 20–100 Hz with 55–65 Hz removed

We tested whether a SEEG pair spectrum predicts another labeled session better than its electrode shaft mean. The test used public OpenNeuro ds005624 version 1.0.0. Seven participants supplied 17 records and ten comparisons. Equal pair names were assumed to identify the same physical contacts. The band was 20–100 Hz with 55–65 Hz removed. Each log spectrum was centered to remove overall power. Pair prediction reduced mean squared error by 51.4 percent. All seven participants improved. The participant bootstrap interval for the error reduction was 0.01349 to 0.02547. After fitted slope removal, copying the remaining shape increased error by 15.0 percent. An exact error split showed that fitted slopes accounted for the primary gain. Copying only pair slopes reduced full shape error by 55.2 percent in an exploratory calculation. A further exploratory check copied half the remaining shape and reduced its error by 11.8 percent. Pairs without shared contacts retained the primary gain. Reversed excerpts gave a weaker, less consistent result. The small selected sample, unknown session dates, and assumed contact identity limit the finding. The package preserves code, spectra, fixed source versions, detailed errors, and reproduction commands.

stereoelectroencephalography · SEEG · spectral prediction · power spectra · session repeatability · log-frequency slope · electrode contacts