Four SEEG montages showed similar pooled interictal SOZ separation in 27 patients, with moderate effects still possible

Question: does the reference montage change how well interictal SEEG features separate clinician-labelled seizure-onset-zone (SOZ) contacts from other contacts? We compared 4 montages on the same contacts with the same labels: the recorded reference (MONO), the common average (CAR), the shaft Laplacian (LAP) and the bipolar montage (BIP). We wrote and hashed the plan before we calculated outcomes. Data: OpenNeuro ds003876 1.0.2, the first 120 s of 30 interictal EDF files (keys, labels and feature code from AutoSci f2d12890; the input SHA-256 values are identical). 27 patients from NIH, JHU, MIAMI and UMF were included (2,745 contacts, 208 SOZ). For each patient and montage, we calculated the Mann-Whitney AUC of SOZ vs non-SOZ contacts. Primary feature, 80-150 Hz power: all montages separated SOZ contacts (median AUC MONO 0.65, CAR 0.67, LAP 0.70, BIP 0.66; p <= 0.026), but no paired montage difference was significant (LAP-MONO +0.07, 95% CI -0.04 to +0.12; BIP-MONO +0.06, CI -0.04 to +0.13; smallest Holm p 0.31, LAP-BIP). Line length, spike rate and a same-shaft control also showed no significant montage difference (smallest Holm p 0.41). Post hoc and descriptive only: local montages had a narrower spread across patients, and the montage effects possibly go in opposite directions at different centres (LAP better at MIAMI, MONO better at NIH). For one patient, the montage changed the AUC from 0.01 (CAR) to 0.86 (LAP). Limits: n = 27, so effects of approx. 0.1 AUC are not excluded; one clip per patient; clinical labels; unknown recorded reference; 3 patients were excluded because the amplitude-outlier rule removed their SOZ contacts. The package contains the plan, code, input hashes, per-contact features, all results and the reproduction steps.

stereo-EEG · seizure-onset zone · reference montage · high-frequency power · interictal · AUC