Spindle phase locking transfers within records but varies across 9 epilepsy patients

We tested if NREM spindles occur at a consistent phase of the local slow oscillation (SO) in SEEG. Data: OpenNeuro ds003498, 9 patients, 5 min NREM each, 462 adjacent bipolar pairs, 7438 spindles from the offline detector of 307cb98b. For each pair with 10 or more spindles (280 pairs), the mean resultant length of the 0.5-2 Hz phase at spindle peaks was compared with 200 circular time shifts. Locking was above the null in 9/9 patients (sign-flip p = 0.002) and in 44% of pairs (5% expected). A preferred phase learned on one half of the record caught 34-47% of the spindles in the other half in a 90 deg window (chance 25%; 9/9 patients, p = 0.002). Locking was stronger at high SO amplitude (8/9). The preferred referential phase was not consistent across patients (Rayleigh p = 0.31). A post-hoc control removed spindles on sharp transients (31% of spindles). Locking stayed in 9/9 patients, but the effect decreased by about one half (held-out hit rate 27-40%). For spindle-coupled closed-loop stimulation, these data support a per-patient, per-pair phase calibration, not a fixed population phase. Limits: offline zero-phase filters (upper bound for real-time use), no IED marks, unknown reference, 5 min per patient.

sleep · slow oscillation · spindles · SEEG · phase locking · closed-loop stimulation · epilepsy · calibration