Interictal SEEG features ranked clinician-marked seizure onset channels in 34 HUP patients, but did not distinguish surgical outcome

In 34 patients with stereo-electroencephalography (SEEG) recordings from the HUP iEEG dataset, each of seven pre-specified features from the first 120 seconds of one interictal clip ranked clinician-marked seizure onset zone (SOZ) channels above other bipolar channels (mean per-patient AUC 0.640–0.753; Holm-adjusted permutation p = 0.0003–0.0007). A composite of transient rate, high-gamma power, and delta power had mean AUC 0.753, median AUC 0.785, and AUC above 0.5 in 32 of 34 patients. AUCs for resected-channel labels showed the same pattern with smaller values. No feature differed between seizure-free (n = 20) and failed-surgery (n = 14) patients after Holm correction. For the SOZ composite, mean AUC was 0.790 versus 0.701 (uncorrected p = 0.084). A nominal interim outcome difference for transient rate in 17 patients did not hold in the full sample. These within-patient rankings do not establish a cross-patient classifier. Limits include one short clip per patient, uncontrolled sleep or wake state, sampling rates of 256–1024 Hz, an unvalidated transient detector, no anatomical correction, clinician labels that are not ground truth, and three SEEG patients not analysed because of a missing file or the run time limit. The analysis used OpenNeuro ds004100 snapshot 1.1.3.

stereo-electroencephalography · epilepsy · seizure onset zone · interictal · bipolar channels · AUC · permutation test · OpenNeuro