Background spectra did not improve calibration-based contact ranking in 9 ECoG patients

We re-analyzed the per-contact table from SCI 46907ea2 to ask whether adding nine task-free background spectral features to 2, 5, 10, or 20 calibration trials improved ranking of visually responsive contacts, and whether an exploratory SEEG background-slope marker was independent of background power. A leave-one-patient-out L2 logistic model was compared with calibration alone using within-patient AUC and two-sided exact sign-flip tests. In 9 ECoG patients, adding background features lowered median AUC at every calibration size: 0.665 to 0.559 (7/9 patients lower, p=0.10), 0.771 to 0.713 (6/9, p=0.18), 0.849 to 0.788 (7/9, p=0.039), and 0.898 to 0.823 (8/9, p=0.051), respectively. The background-only median AUC was 0.518. In 4 SEEG patients, changes were inconsistent and all p values were at least 0.25. Within each SEEG patient, slope and mean 13–170 Hz background power correlated at r=-0.83 to -0.91. The slope AUC fell to 0.58, 0.45, and 0.39 in three patients after removing four band powers; it was not computed for the fourth patient, which had 6 channels. The exploratory slope marker therefore mostly shared information with low broadband power. Calibration and slope AUCs reproduced the earlier analysis. This re-analysis uses the same data and labels, not an independent sample. It includes 9 ECoG and 4 SEEG patients, one cross-patient combination method, four uncorrected calibration sizes, and clinical epilepsy implants; a within-patient Bayesian prior could behave differently. There was no anatomy analysis.

electrocorticography · stereo-EEG · brain-computer interface · contact selection · calibration · background spectrum · visual responses · area under the curve · OpenNeuro ds004194