In 36 HUP patients, seizure-onset-zone pairs were net sources of bivariate Granger influence, but the effect did not persist after power correction

We tested whether seizure-onset-zone (SOZ) bipolar pairs receive more directed influence than they send in short interictal SEEG segments from the HUP iEEG Epilepsy Dataset (OpenNeuro ds004100, version 1.1.3). The analysis included 36 patients, 3,687 bipolar pairs, and 392 SOZ pairs, using 20–90 seconds per patient. The primary measure was mean bivariate Granger causality into a pair minus mean causality out to pairs on other shafts. Mean per-patient AUC for SOZ versus non-SOZ pairs was 0.443 (10/36 patients above 0.5; two-sided sign-flip p=0.014, Holm-adjusted p=0.028), opposite to the net-sink hypothesis. Higher out-strength (AUC 0.576, p=0.013), rather than lower in-strength (AUC 0.547, p=0.119), accounted for the direction. After per-patient linear residualization on log broadband power, net AUC was 0.483 (Holm-adjusted p=0.451). Net strength was negatively correlated with log power (mean Spearman rho −0.17; 30/36 patients negative), so the net-source effect could not be separated from the higher broadband power of SOZ pairs. Granger causality is invariant to multiplying a channel by a constant, so this association is not a gain artifact; signal-to-noise ratio or spectral shape may contribute. Time-reversed net had AUC 0.526; the secondary time-reversal-corrected measure had AUC 0.453 (p=0.041, not corrected for multiple comparisons). Each patient contributed one short segment; sleep stage and spikes were not controlled, the bivariate model used fixed order 8, SOZ labels were clinical, power residualization was linear, and all 36 patients came from one center. These results do not support the SOZ net-sink hypothesis in these data, and the observed net-source effect does not persist after power correction.

seizure-onset zone · stereo-EEG · Granger causality · epilepsy · directed connectivity · broadband power · source-sink hypothesis