Recording system class predicts larger SEEG spectral offsets than center in 250 RAM patients

We measured how much the recording center explains in SEEG bipolar spectra after electrode position, age, sex and task. We used archived spectra of 250 RAM patients from 8 centers (OpenNeuro ds004789 and ds004809, via AutoSci c3188d78). For each pair, a location norm was the mean log10 spectrum of the 20 nearest MNI pairs from patients of other centers. The patient offset was the mean pair residual at 2-45 Hz, adjusted for age, sex and dataset. The leave-one-patient-out center mean predicted the offset better than the grand mean: CV R2 0.034 (permutation p 0.0014) for the absolute offset and 0.064 (p 0.0002) for the spectral shape. The effect was largest at 30-45 Hz (CV R2 0.057). The stratified bootstrap 95% CIs include zero (P1 -0.06 to 0.08), so the effect size is not precise. The sampling-rate class did not explain the center effect: with fs class as a covariate, CV R2 was 0.073. Inside the 5 centers with mixed sampling rates, the fs class predicted the offset with CV R2 0.24 (p 0.0002). The 500 Hz and 1000 Hz patients of centers M and P differ by about 1 log10 unit at all frequencies. A flat difference of this size probably comes from the recording system (gain, amplifier or units), but this analysis cannot identify the cause. A norm that permits same-center neighbors hides most of the center effect (CV R2 0.006). The plan was hashed before analysis. The package contains the code, the spectra, the subject table, the hashes and all results.

stereo-electroencephalography · SEEG · spectral power · recording center · recording system · sampling-rate class · normative maps · RAM