Background spectra did not rank ECoG contacts; a flatter slope ranked SEEG channels in 4 patients (exploratory)

We tested whether pre-stimulus background spectra from one run can rank visually responsive intracranial EEG contacts in another run, and compared this with rankings from a few calibration trials. The analysis used 9 epilepsy patients (p03–p11) in OpenNeuro ds004194, with ECoG and SEEG recordings sampled at 512 Hz. Labels came from run 2; features came only from the 0.5 s before each stimulus in run 1. A leave-one-patient-out logistic model did not rank ECoG contacts: the median within-patient AUC was 0.52 (5 of 9 above 0.5; exact p = 0.37). Rankings from the first 2 run-1 trials had a median AUC of 0.67 (9 of 9 above 0.5; p = 0.002); 5, 10, and 20 trials gave 0.77, 0.85, and 0.90. For SEEG, only 4 patients met the minimum response-count criterion. The full background model had median AUCs of 0.56–0.62. A flatter 30–170 Hz background slope ranked responsive channels with AUCs of 0.78, 1.00, 0.79, and 1.00 (median 0.89), while responsive channels also had lower 13–170 Hz background power. The slope was one of 5 SEEG feature sets tested, and the small sample gives a minimum possible exact p of 0.0625; this result is exploratory. The background windows were inter-trial periods within a task, not rest. Residual late responses may remain. No anatomy or tissue labels were used, and the recordings came from patients with epilepsy and clinical implants. The package preserves analysis code, per-contact results, source-file hashes, the figure, and reproduction instructions. The download script queries a live public S3 mirror, so the exact immutable dataset release used is uncertain.

intracranial EEG · ECoG · SEEG · visual cortex · brain-computer interface · calibration · background spectrum · AUC · neural signal processing