In OpenNeuro ds003876, SSI did not outperform 1–45 Hz power in identifying clinical seizure-onset channels

We re-implemented the interictal source-sink index (SSI) from the published description and compared it with log 1–45 Hz power in 30 patients from four centers in OpenNeuro ds003876. We analyzed one 60-s interictal segment per patient (3,270 good SEEG channels, including 216 clinical seizure-onset-zone [SOZ] channels). In the prespecified, per-window z-scored analysis, mean per-patient SSI AUC was 0.31 (95% CI 0.24–0.38; 5/30 patients above 0.5; Holm-adjusted p = 0.00015), indicating lower SSI at SOZ channels. Power AUC was 0.73 (0.66–0.80), and the mean SSI-minus-power AUC difference was −0.42 (−0.53 to −0.31; Holm-adjusted p = 0.0001). In a post hoc analysis without z-scoring, SSI AUC was 0.58 (0.51–0.64; p = 0.031), but remained below power and showed no SOZ separation after linear power control (residual AUC 0.46, p = 0.32; the z-scored residual AUC was 0.46, p = 0.28). The SSI–power correlation changed from −0.46 (−0.56 to −0.34) to +0.45 (0.33–0.56) without z-scoring, so the direction of the SSI effect depended on normalization. This re-implementation did not reproduce the reported AUC of about 0.77. Interpretation is limited by unspecified implementation details, one arbitrary 60-s segment per patient, clinical SOZ labels, and less well-determined least-squares matrices in some windows. The source files, code, results, segment hashes, URLs, and other provenance are preserved in the package.

source-sink index · intracranial EEG · seizure onset zone · epilepsy · signal power · OpenNeuro · replication