A causal sleep spindle trigger fired 0.51 s after onset and left 0.18 s in a 71-subject SEEG replication

In an offline simulation of 240 s of NREM sleep per subject from 71 independent SEEG subjects, a causal 11–16 Hz RMS spindle trigger with 100 ms dwell fired a median 0.51 s after the offline spindle onset (95% CI 0.51–0.53), leaving 0.18 s (0.16–0.22); the earlier study reported 0.52 s and 0.21 s. Across 69 subjects and 1,048 bipolar pairs, precision was 0.40 (0.33–0.43) and sensitivity was 0.63 (0.60–0.67), compared with 0.40 and 0.70 in the earlier study. Time left was below 250 ms in 53 of 69 subjects (sign test p = 9e-6). Increasing dwell from 0 to 200 ms reduced sensitivity from 0.80 to 0.42 in 67 of 68 subjects (p = 5e-19), without increasing precision (0.41 to 0.39, p = 0.056). Spiking or seizure-onset-zone pairs had more offline spindles (+1.4 per min; 38 of 50 subjects, p = 3e-4), while trigger latency, time left, and precision did not change. This replication used threshold-based offline reference events and does not establish online performance; the HFO analysis was not repeated because these datasets have no HFO marks.

sleep spindles · SEEG · causal detection · intracranial EEG · replication · signal processing