Stimulation-site information persisted across 10–300 ms and transferred across artifact shapes in one patient, without proving a neural source

In one patient and one run from OpenNeuro ds003708, 12 unstimulated SEEG contacts identified which of 37 cortical sites received a 6 mA pulse in every post-pulse window tested: top-1 accuracy was 45.2% at 10–50 ms, 28.6% at 50–100 ms, 30.2% at 100–200 ms, 25.7% at 200–300 ms, and 37.6% at 10–300 ms (chance 2.7%). The pre-pulse control accuracy was 1.1%. At eight sites with two sampled shapes of the 0–1.5 ms artifact, templates from the other shape classified pulses at 36.5% accuracy in the 10–300 ms window, compared with 32.9% for same-shape templates (chance 12.5%; permutation p = 0.005, the smallest possible with 200 permutations). The artifact shapes were probably two sampling phases of a 0.2 ms biphasic pulse on a 0.49 ms sample grid, rather than opposite current directions. These results show that site information continued to 300 ms and did not depend on the sampled artifact shape. They do not establish a neural source: slow amplifier or electrode recovery could explain the late signal, and no current-reversed pulses were labeled. The study includes one patient, one current, one run, and only eight sites with both artifact groups. The analyzed source was fetched from an unversioned OpenNeuro S3 URL, so its exact dataset release is unknown.

stimulation-site decoding · SEEG · intracranial EEG · electrical stimulation · stimulation artifact · latency · template correlation · OpenNeuro