Contiguous null labels inflate per-patient SEEG seizure-onset-zone false-positive rates
Archived interictal SEEG feature tables were used to test whether per-patient Mann–Whitney tests of seizure-onset-zone (SOZ) versus non-SOZ bipolar pairs remain calibrated when null labels share the SOZ's contiguous shaft structure. In the primary HUP and ds003876 cohorts (58–59 patients per feature), spatial sham labels produced false-positive rates of 0.181–0.242 at alpha 0.05, compared with 0.047–0.048 for exchangeable random labels. Holm-adjusted p values were 0.0004 for all four primary features, and median AUC variance inflation was 1.76–2.29. The result persisted in a sensitivity analysis that used non-SOZ pairs only from shafts without SOZ pairs. Secondary cohorts showed spatial false-positive rates of 0.146–0.152 for ds005398 and 0.066–0.130 for ds007118–20. For real SOZ labels in the primary cohorts, 99 of 235 patient-feature tests were significant with the naive test and 62 with spatial permutation. These results concern per-patient p values, not bias in group-mean AUC. Limitations include clinical SOZ labels, short selected interictal clips, adjacency inferred from pair names without coordinates, possible dependence across shafts, and the normal-approximation Mann–Whitney test.
SEEG · epilepsy · seizure-onset zone · Mann–Whitney test · false-positive rate · spatial permutation · interictal features · statistical calibration
References
6- usesIn 18 UCLA pediatric SEEG patients, 80–250 Hz bursts and tonic levels both marked clinical seizure-onset channels; neither outperformed the other sci:sha256:2e8119ca326b41f7ea7d9203153c75297ce6512a42fe05ac4177e59632e4f008
ds005398 pair table (tonic, meanpow, SOZ) is a secondary cohort.
- supportsNearby SEEG bipolar pairs have more similar spectral parameters than distant pairs in epilepsy patients sci:sha256:0db79463de7725c2bb080901723abeacfccd2e488a130ef3fb47614409f9ac30
That work reported spatial autocorrelation of pair features along shafts. We show its consequence: contiguous null labels give 3.8-5.2 times the nominal false-positive rate in per-patient SOZ tests.
- usesThe 30–45 Hz aperiodic exponent did not separate seizure-onset channels in 31 SEEG patients sci:sha256:970015c7ad657e77ac8f470146719405ea5536dba9b4dabf2cdf20f1f8c23615
HUP (ds004100) pair-level interictal features and SOZ labels are the primary cohort.
- usesIn 117 SEEG patients, the top interictal HFA pair was in the clinical SOZ less often than its AUC predicted sci:sha256:6c85dc6f3540bdd41aaeb79a7fbecee27e8b50ac476c990e8241c271ddd4e414
We copied the four archived feature tables and their SHA256SUMS from the inputs folder of this publication.
- usesSlow-wave–HFA coupling and HFA power distinguish clinical SOZ depth pairs in 43 patients sci:sha256:9f9a03f5ae81593b8a29528c314db6efb6a4c1b7fb63e4d09e778c7b70330fe7
ds007118-20 pair table (log_hfa, log_delta, SOZ) is a secondary cohort.
- usesAcross 29 SEEG patients, the 30–45 Hz exponent did not separate SOZ channels, while four other features did sci:sha256:115bc8cb0b4c9e536b5f083843a2745508d3c6635a8156b661a9e084f51582a7
ds003876 pair-level interictal features and SOZ labels are the second primary cohort.
Cited by
3 publicationsPublic catalog entries on Sciscape.
- usesWithin-shaft SEEG features identified SOZ pairs for three markers but not the aperiodic exponent
That work showed that SOZ labels are contiguous on shafts and per-patient tests are not calibrated. We used it to design a shaft-matched comparison with the patient as the unit.
- usesClinical SOZ pairs did not lead interictal spikes in 27 SEEG patients (OpenNeuro ds004100; mean rank difference -0.020, 95 % CI -0.072 to 0.034)
Its finding that per-patient SOZ tests have 0.18-0.24 false positives under contiguous-block null labels made us not interpret per-patient p-values.
- usesContiguous shaft-block sham labels inflate pooled-pair SOZ false positives in four SEEG cohorts
We reuse its block-placement null code without change and its four input tables. We extend its per-patient result to cohort-level tests.