Across 15 SEEG patients, causal ecHT reduced offline theta-phase error while simulated pulse delay increased it

We evaluated causal phase estimation for phase-locked stimulation in session 1 of all 15 patients in OpenNeuro ds004752 v1.0.1, using bipolar SEEG pairs from artifact-free 8 s trials resampled to 250 Hz. The analysis plan was written before results were viewed. We compared causal estimators using a zero-phase filtered reference and treated patients as the unit of inference, with exact two-sided sign-flip tests. For theta (4–8 Hz), endpoint-corrected Hilbert transform (ecHT) had a median error of 53.7° versus 61.7° for a naive causal filter; ecHT error was lower in all 15 patients (p=6.1×10⁻⁵). An amplitude gate and bias calibration on odd trials, evaluated on even trials, reduced median error to 34.6° with a 74% hit rate (absolute error <45°). In fixed-delay offline simulations, gated and calibrated error increased by 7.8° at 25 ms and 15.4° at 50 ms; the increase occurred in all 15 patients (p=6.1×10⁻⁵). The hit rate declined from 74% at 0 ms to 66%, 59%, and 43% at 25, 50, and 100 ms. Instantaneous-frequency phase advance reduced error by 2.2° at 50 ms and 1.8° at 100 ms relative to a fixed 6 Hz advance. Delta-band delay costs were approximately half the theta costs. Hippocampal and non-hippocampal pairs showed no significant difference (median difference −0.6°, p=0.22); this result does not establish equal accuracy. The study is an offline simulation with fixed delay, no jitter or stimulation artifact, and error measured relative to an estimated filtered reference. It covers one verbal working-memory task, one center, epilepsy patients, and session 1 only.

SEEG · phase estimation · phase-locked stimulation · endpoint-corrected Hilbert transform · theta · delta · pulse delay · epilepsy · offline simulation