Three response directions gave lower AUROC than full distance for excluded stimulation shafts in five SEEG sessions

Distance along three fitted response directions did not improve detection of stimulation shafts excluded from calibration. We used OpenNeuro ds004457 version 1.0.2. The test included five SEEG sessions, 183 sites, 38 shaft exclusions, and 732 distinct test pulses. The first three known pulses per site fit templates, covariance, and directions. Separate fourth pulses set cutoffs. Final four pulses tested all scores. Equal-patient mean AUROC was 0.5874 for three directions and 0.6038 for full Mahalanobis distance. The paired change was -0.0164 and was negative in four patients. Its descriptive 95% patient bootstrap interval was -0.0287 to -0.0012. The exact sign-flip reference was p = 0.1875. Three directions retained 88.7% of known template variation on average. This compression did not establish better detection of excluded sites. All six tested score gates accepted most excluded pulses. The package preserves compact voltages, pinned source retrieval, models, predictions, three figures, and verification code. All source hashes passed. A raw rebuild matched every input array and event table. Independent formulas and a separate process reproduced the results. This reused offline benchmark does not establish online performance or a population effect.

stereoelectroencephalography · electrical stimulation · brain-computer interface · Mahalanobis distance · out-of-distribution detection · AUROC · calibration