In three SEEG patients, most unstimulated pairs showed no detectable difference from 50 ms after logged stimulation offset

A BCI that records after it stimulates must know how long the recording stays changed. We measured this in OpenNeuro ds005491 (RAM catFR2, open-loop encoding stimulation, 500 Hz monopolar EDF). We used 3 SEEG patients (R1031M, R1074M, R1180C), one stimulation site each, 50 Hz trains of 4.6 s at 1.0-1.5 mA. For each patient, we took 20 trains and 17-20 control epochs from no-stimulation lists, matched by word serial position. We read the data by HTTP byte range. We made adjacent-contact bipolar pairs and removed pairs with a stimulated contact (166 pairs, 8-101 mm from the site). In 100 ms windows (50 ms step) for 1 s after the logged train end, we compared 70-150 Hz log power and the voltage offset from the pre-train mean with permutation tests and BH-FDR. In the first window (0-100 ms), 3 of 11 pairs (R1031M) and 55 of 68 pairs (R1180C) had higher 70-150 Hz power. In R1180C, these pairs were up to 101 mm from the site, so this is probably a widespread artifact or residual train, not a local effect. R1074M had no significant pair. From the second window (50-150 ms) on, 2 of 2988 tests passed q<0.05, both single isolated windows. Thus, a blanking time of about 100 ms after the logged train end was sufficient at most pairs for these measures. Limits: the nearest pair in R1074M (7.8 mm) and R1180C (8.4 mm) had a mean voltage offset of 1-5 pre-train SD for the full 1 s that did not pass the test because the spread was large. Thus, a sustained offset within about 10 mm of the site is not excluded. The power is low (3 patients, about 20 epochs per group), the band stops at 150 Hz, and the train end comes from events.tsv, not from the signal. The package has numpy-only code, the SVG figure, per-pair tables, and the file list with SHA-256 values.

SEEG · open-loop stimulation · post-stimulation recovery · stimulation artifact · high-frequency power · blanking time · brain-computer interface