In 256 RAM SEEG patients, a patient's distant pairs predicted 2–45 Hz spectra better than location-matched pairs from other patients

Question: does a location-matched norm predict the spectrum of a SEEG bipolar pair better than the same patient's distant pairs? Data: archived spectra of AutoSci c3188d78 (OpenNeuro ds004789 3.1.0 and ds004809 2.2.0; 256 patients; 27957 pairs; 60 s from the first word event). Method: a plan was written before the analysis (SHA-256 recorded). For each pair, we predicted the 2-45 Hz log10 spectrum from: the 20 nearest MNI pairs of other patients (L); the mean of the same patient's pairs 20 mm or more away (P); a grand mean of other patients (G); and L plus a patient offset from the distant pairs (LP). The patient is the unit. We used sign-flip tests and Holm correction. Results: own distant pairs had 33 % less error than L for the absolute spectrum and 19 % less for the spectrum shape (153/256 and 162/256 patients; Holm p 0.0002). L had 8 % less shape error than G. LP had 39 % less absolute error than L. Patient means explained 46 % (absolute) and 35 % (shape) of the variance; the location norm explained 5-6 %. A norm from other-site patients only was 1.5-3.7 % worse. Near 8.5-11 Hz, location and patient predicted the shape about equally. Results did not change with k = 10 or 40, a 30 mm threshold, or each dataset alone (T1 was not significant at k = 10). Limits: epilepsy patients, one task segment, no tissue or seizure-onset labels, MNI midpoints only, and own pairs can be on the same shaft. Hardware and implant strategy can cause part of the patient effect. Normative SEEG spectral maps should include a per-patient calibration.

SEEG · intracranial EEG · spectral power · normative maps · patient identity · MNI · memory task · epilepsy · bipolar pairs · spectral shape