In 36 HUP SEEG patients, the source-sink index marked SOZ pairs but added no information beyond broadband power
We evaluated the interictal source-sink index (SSI) in 3,687 bipolar SEEG pairs from 36 patients in OpenNeuro ds004100. The analyzed clips were 20–90 s per patient. We used a linear time-varying model with 1-s ridge windows at 256 Hz and compared within-patient SOZ discrimination by per-patient AUC. SSI AUC averaged 0.669 (30/36 patients above 0.5; sign-flip p = 1e-5, Holm-adjusted p = 2e-5). Log broadband power alone had mean AUC 0.694. SSI and power were strongly correlated within patients (mean Spearman rho 0.88, range 0.76–0.95). After pre-specified within-patient linear residualization on log broadband power, SSI AUC was 0.490 (p = 0.78; Holm-adjusted p = 0.78), so the pre-specified test found no added SOZ information beyond power. With channels z-scored to unit variance, SSI AUC was 0.407 (p = 0.008). These findings apply to this implementation and dataset; the analysis used short interictal clips, clinical SOZ labels, and a montage and model that differ from the original publication, and is not a direct replication.
source-sink index · stereo-EEG · seizure onset zone · broadband power · epilepsy · AUC · OpenNeuro ds004100
References
3- supportsInterictal SEEG coherence marks SOZ-to-SOZ links but does not improve power models in 36 HUP patients sci:sha256:cd0d34ab255f0a3cc5fc470aef3491bd7a742a7bd3459508319ba8152dc6a9f3
On the same segments, a second network measure (LTV source-sink index) also does not add SOZ information to a power model.
- usesInterictal power and transient features distinguish SOZ channels after depth matching in 36 SEEG patients, but aperiodic exponent and outcome do not sci:sha256:398ad0c4813dba23a5e3270f2aa74722c599a5916eec0eb90a71accc37c3eed1
We use its fetch code, the same interictal segments (same byte ranges), the same bipolar pairs and SOZ labels, and its EDF reader and AUC code.
- usesdoi:10.18112/openneuro.ds004100.v1.1.3
The publication analyzes interictal SEEG recordings from OpenNeuro ds004100 version 1.1.3.
Cited by
3 publicationsPublic catalog entries on Sciscape.
- supportsIn 36 HUP patients, seizure-onset-zone pairs were net sources of bivariate Granger influence, but the effect did not persist after power correction
A scale-free lagged measure (Granger causality) also does not show the SOZ as a net sink; it shows a net source (AUC 0.443) that cannot be separated from broadband power, in agreement with the power confound and the z-scored SSI reversal found in 3eedcbd6.
- usesIn 36 HUP patients, seizure-onset-zone pairs were net sources of bivariate Granger influence, but the effect did not persist after power correction
Reused fetch_segments.py, the S3 key list, the same EDF byte ranges (SHA-256 identical for all 37 segments), the bipolar montage and the SOZ labelling rule.
- supportsIn interictal SEEG of 36 HUP patients, neural fragility separates SOZ pairs only because it follows broadband power (rho 0.98); after power control its SOZ AUC is 0.43
A second network measure (fragility) on the same data also gives no SOZ information after control for broadband power.
- usesIn interictal SEEG of 36 HUP patients, neural fragility separates SOZ pairs only because it follows broadband power (rho 0.98); after power control its SOZ AUC is 0.43
We use its fetch code, EDF segments, bipolar pairs, SOZ labels, preprocessing and test design. We also use its per-pair SSI values.
- supportsIn OpenNeuro ds003876, SSI did not outperform 1–45 Hz power in identifying clinical seizure-onset channels
In a different dataset (ds003876, 4 centers, CAR, 500-ms windows), we also found that the SSI tracks channel power, that z-scoring reverses its SOZ effect (AUC 0.31 vs 0.58 without z-scoring), and that SSI gives no SOZ information after power control (residual AUC 0.46 in both versions).