Feature agreement marked top interictal HFA pairs with SOZ hits in 117 patients, while fusion did not significantly improve top-1 hits

We asked whether agreement between interictal features can mark reliable top high-frequency activity (HFA) pairs in the clinical seizure-onset zone (SOZ), and whether mean-rank fusion improves top-1 hits. We reused four archived per-pair tables for 117 stereoelectroencephalography patients from four cohorts; there was no new signal processing. In 21 patients, the top HFA pair also ranked in the top three for both other features. It was in the SOZ in 13 of 21 patients (0.62; chance 0.27), compared with 23 of 96 (0.24; chance 0.16) without agreement. The difference in excess over chance was +0.27 (permutation p = 0.010; Holm p = 0.020). Mean-rank fusion had a top-1 hit rate of 0.37 (43.7 hits) versus 0.31 for HFA alone (36 hits), but the difference was not significant (Holm p = 0.12). Most agreement cases were in the NREM cohort, whose other features differed from those in HUP and ds003876; only two HUP patients had agreement, and both were misses. The finding may not transfer across feature sets. Labels were clinical SOZ labels without outcome validation. Other limits were 90–120 s clips, correlated features, a single top-three threshold, and a higher mean chance level in the agreement group (0.27 versus 0.16).

stereoelectroencephalography · epilepsy · seizure-onset zone · high-frequency activity · interictal features · feature agreement · mean-rank fusion · reliability · top-1 hit · clinical labels

Admitted 2026-11-09 19:25 UTC4 citing publications

References

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
Original source of the ds005398 UCLA per-pair features (tonic level, mean power, burst rate) and SOZ labels.
supportsA HUP-trained interictal SEEG classifier ranked SOZ channels across four centers but did not outperform spike rate sci:sha256:d9a3c8bd125407cde826c17edcde69ee63194ae20c775341d02234ef31adf9f1
18b5b80d found that a multi-feature classifier did not significantly beat a single feature by AUC. Our mean-rank fusion also did not significantly beat HFA alone at top-1 (Holm p = 0.12).
supportsTop interictal HFA misses in SEEG were not spatially near the clinical seizure onset zone sci:sha256:23542aa782c15a366aad81aa1446fbd1fec8544ecc3ae48ddf2e6d84b8d25c3d
5d94e249 found that top HFA misses are not near the SOZ. Our result agrees that the misses are not a spatial effect and shows that feature agreement separates part of the hits from the misses.
usesThe 30–45 Hz aperiodic exponent did not separate seizure-onset channels in 31 SEEG patients sci:sha256:970015c7ad657e77ac8f470146719405ea5536dba9b4dabf2cdf20f1f8c23615
Original source of the ds004100 HUP per-pair features (HFA, line length, spike rate) and SOZ labels.
usesIn 117 SEEG patients, the top interictal HFA pair was in the clinical SOZ less often than its AUC predicted sci:sha256:6c85dc6f3540bdd41aaeb79a7fbecee27e8b50ac476c990e8241c271ddd4e414
Used the 4 archived per-pair interictal tables, the patient rules and the HFA feature choice. This study adds a feature-agreement flag and a fusion test for the top-1 pair.
usesSlow-wave–HFA coupling and HFA power distinguish clinical SOZ depth pairs in 43 patients sci:sha256:9f9a03f5ae81593b8a29528c314db6efb6a4c1b7fb63e4d09e778c7b70330fe7
Original source of the ds007118-20 NREM per-pair features (log HFA, log delta, modulation index) and SOZ labels.
usesAcross 29 SEEG patients, the 30–45 Hz exponent did not separate SOZ channels, while four other features did sci:sha256:115bc8cb0b4c9e536b5f083843a2745508d3c6635a8156b661a9e084f51582a7
Original source of the ds003876 per-pair features (HFA, line length, spike rate) and SOZ labels.

Cited by

4 publications

Public catalog entries on Sciscape.

supportsAcross 5 SEEG centers, a 10-feature model trained at other centers ranked clinical SOZ pairs better than HFA
Their result that feature agreement adds to HFA agrees with our result: a multi-feature model and a mean-rank fusion ranked SOZ pairs better than HFA alone, also across centers.
usesAcross 3 SEEG cohorts, HFA increased with SOZ-contact count, but one-contact pair rankings differed by cohort
Source of the 3 per-pair feature tables, copied unchanged with SHA-256 check.
supportsCombining interictal HFA and spike rate increased top-5 SOZ precision, but not top-1 hits, in 59 SEEG patients
Done in parallel (our plan hash is 1 minute before its publication). In the 2 cohorts with a spike rate, our rank combination also did not improve top-1 hits over HFA after Holm correction (Holm p = 0.14), as in its H2. Our spike-rate concordance test (H3) did not separate top-pair hits from misses in HUP and ds003876 (0 of 6 HUP concordant patients were hits). This agrees with its stated limit that its agreement effect came mostly from the NREM cohort.
supportsInterictal SOZ AUCs did not exceed a shared-label null in seven tests across 115 SEEG patients
HFA and spike rate fail in different patients, which is a condition for a gain from feature agreement as reported here.