Pulse variance correction reduced ranking from two pulses in four of five SEEG patients on five separate test pulses

We tested whether pulse variance correction improves SEEG response ranking with two pulses. We used public OpenNeuro ds004457 version 1.0.2. The test included 1,080 stimulation-site and receiving-contact pairs from five patients. Each method selected the nominal top 10% of pairs. Five separate pulses set the test ranks. We averaged all 2,520 disjoint pulse allocations within each patient. Mean rank AUC was 0.9259 for mean-waveform energy and 0.9257 after correction. The change was negative in four patients and positive in one. The patient mean change was -0.00024; the exact two-sided sign-flip p value was 1.00. The empirical patient bootstrap limits were -0.0160 to +0.0230. These limits do not show equivalence. Correction reduced rank AUC in all five patients at a 20% selection quota. It raised mean AUC when only post-pulse energy was used. A normal pulse model shows that before-pulse subtraction can remove energy bias when noise variances match. In that case, the extra correction adds variance. The patient results do not establish this model. The package includes code, windows, hashes, all allocations, checks, and a raw retrieval recipe. All 3.48 million saved waveform bins matched the source recordings exactly. The fixed sample tests measurement ranks. It does not establish anatomical connections or clinical accuracy.

SEEG · intracranial electrical stimulation · pulse variance correction · response ranking · rank AUC · OpenNeuro