In three NREM-sleep SEEG patients, bipolar SO phase prediction was weak even at 0–10 mm and error rose with distance (not significant)

We measured whether the 0.5–2 Hz slow-oscillation (SO) phase at one stereoelectroencephalography (SEEG) bipolar channel predicts phase at another channel during NREM sleep. This matters for closed-loop stimulation when the sensing and target contacts differ. We analyzed one 120 s segment per patient from OpenNeuro ds007119 (two patients) and ds007120 (one patient), using HTTP byte-range reads from the OpenNeuro S3 mirror. Three patients met the rule of at least 12 SEEG contacts with coordinates and at least 8 adjacent bipolar channels. After zero-phase 0.5–2 Hz filtering, Hilbert phase estimation, and gating on sensing-channel SO envelope above its median, we calculated mean absolute phase error after a fixed circular-mean offset correction fitted in-sample. Across 7,862 ordered channel pairs, median errors at 0–10 mm were 68.5–70.6° (median phase-locking value 0.26–0.29); errors increased with distance in all three patients (Spearman rho 0.21, 0.22, 0.54). With n=3, the group test was not significant (minimum possible exact two-sided p=0.25). At matched 10–30 mm, same-shaft pairs had slightly lower median error than different-shaft pairs in two patients; no test was performed. The in-sample offset makes these errors lower bounds for a real system. Limits include three patients, one 120 s segment each, no SO-event detection or within-segment sleep-stage check, and bipolar rereferencing that removes shared widespread SO activity. For bipolar SEEG, the findings suggest sensing at or near the target contact. Exact dataset release identity is uncertain; source URLs, ETags, byte ranges, and record indexes are preserved.

slow oscillation · SEEG · bipolar channels · phase prediction · NREM sleep · closed-loop stimulation · phase error · OpenNeuro