Median offsets from other shaft groups reduced mean SEEG power ratio prediction error before seizure onset in 26 selected patients

We tested a baseline adjustment with public SEEG data from OpenNeuro ds004100 version 1.1.3. The analysis reused 2,742 bipolar pairs from two seizure records in each of 26 selected patients. Each target shaft name group was excluded from its own offset. A separate window before onset tested the power ratio prediction. Mean absolute error fell from 3.146 to 2.570 dB. The equal-patient mean reduction was 0.576 dB, with a 95% patient bootstrap interval of 0.211 to 1.018 dB. Error fell in 16 patients and rose in 10. Local calibration had a lower mean error of 1.945 dB. The method needs target calibration signals from other groups. Agreement with a local 6 dB rule also increased. These comparisons do not establish clinical detection accuracy. The local plan was fixed before the new calculations, after the earlier result in the same cohort was read. This is an informed secondary analysis. The package preserves code, derived data, patient results, sampling arrays, a raw example, source hashes, pinned retrieval recipes, and reproduction steps. Fresh source extraction and independent calculations checked the results.

SEEG · baseline calibration · power ratio · signal processing · seizure onset