Subtraction of an external RAM SEEG band-power map decreases delta SOZ AUC in HUP interictal SEEG (36 patients, 10 mm MNI kernel)

A normative map of within-patient centered log band power (1–30 Hz) was built from 256 task SEEG recordings from the RAM cohort (27,974 bipolar pairs) using Gaussian kernel regression on MNI positions, then applied to interictal SEEG from 36 HUP patients. The map transferred to HUP non-SOZ pairs, with mean per-patient correlations of 0.46–0.49 and R² of 0.15–0.19 (all Holm-adjusted p < 0.001). For delta power, subtracting the map reduced mean SOZ AUC from 0.724 to 0.688 (difference −0.037, 95% CI −0.066 to −0.010, p = 0.012); 25 of 36 patients had lower AUC after subtraction. The position-based prediction alone had AUC 0.699 (p = 0.001). Theta showed a smaller difference (−0.025, Holm p = 0.074); alpha and beta showed no change. Run-02 replicated the delta result (difference −0.029, p = 0.034), as did 5 mm and 20 mm kernels. A multiband absolute-deviation score did not separate SOZ pairs (AUC 0.53). Thus, the external map transferred between cohorts but did not improve interictal SOZ discrimination; part of the raw delta association was related to electrode position, while the deviation retained SOZ information (AUC 0.69, p < 0.001). The RAM map includes people with epilepsy and task periods, may include epileptic tissue, and the deviation used a fixed coefficient of 1. HUP and RAM data used different pipelines, registration error was not measured, the sample included 36 patients, and clinical SOZ labels are a proxy for the epileptogenic zone. Outcome was not used. The analysis was preregistered.

stereo-EEG · SEEG · seizure onset zone · SOZ · normative map · band power · delta power · MNI · epilepsy · AUC · cross-cohort transfer