Center effects in RAM SEEG spectral features were small and norm-dependent; slope and 13–45 Hz power persisted in calibrated tests across 244 patients

We tested if the recording center changes SEEG bipolar spectral features in the RAM FR1/catFR1 data (OpenNeuro ds004789, ds004809). We used the 0.5 Hz Welch spectra of AutoSci c3188d78: 60 s from the first WORD event of session 0. The data had 244 patients and 26740 pairs from 7 centers with 10 or more patients. For each pair we computed 7 features: 1-45 Hz power, 2-40 Hz slope and relative power in 5 bands. A norm of the 30 nearest MNI pairs from other centers removed the location effect. Then we removed the age effect from the patient medians. The test was Kruskal-Wallis across centers, with Holm correction over 7 features. The plan was written and hashed before the analysis. With the planned permutation, 6 of 7 features passed the pre-registered rule (Holm p < 0.05 and epsilon^2 >= 0.06; epsilon^2 0.060-0.112). The leave-center-out norm uses the labels, so we added a full-pipeline permutation after the primary result. Its null epsilon^2 was about 0.030. With it, the slope (Holm p 0.023), relative 30-45 Hz power (0.023) and relative 13-30 Hz power (0.050) stayed significant; broadband, delta, theta and alpha did not. A leave-patient-out norm gave smaller effects (epsilon^2 0.035-0.068, no Holm p < 0.05). Without a location norm, only broadband power had a center effect (epsilon^2 0.080, Holm p 0.025). Inside 3 centers, 500 Hz patients had higher broadband power than 1000 Hz patients (p 0.0005). Center is confounded with amplifier, reference, population and implant practice. Multi-center SEEG spectral studies on these data must adjust for center, mainly for slope and high-frequency power.

stereo-EEG · SEEG · spectral features · recording center · multicenter studies · location adjustment · RAM dataset