An excess variance rule lowered prediction loss for HFO counts in 135 excluded SEEG sleep records from seven patients under record exchangeability
We tested prediction of released HFO overlap counts in 135 five-minute SEEG sleep records from seven patients. Each complete record was excluded before fitting. A fixed rule for excess count variance was compared with a constant Poisson rate rule that included rate estimation uncertainty. The rules used only other records from the same patient. Mean negative log prediction probability fell from 3.027 to 1.851 natural log units per channel per record. Each patient had equal weight. Loss was lower in all seven patients. The mean difference was -1.176. The exploratory 95 percent patient bootstrap interval was -1.999 to -0.512. The exact two-sided sign-flip test gave p = 0.015625 under independent patient effects and null sign symmetry. Count interval coverage rose from 84.41 to 94.47 percent. The alternative intervals contained 97.93 percent model probability, so a calibration gap remained. The result applies to excluded records under exchangeability within this source sample. It does not test later-time prediction, detector accuracy, or surgical outcomes. The package preserves source annotations, versions, hashes, code, every prediction, checks, and reproduction instructions.
high frequency oscillations · stereoelectroencephalography · count prediction · negative binomial model · record exchangeability · epilepsy
References
6- usessci:sha256:73e42f2543dcded2b2577ecbc89c09eac8e76acdb88dd38622ed338522d5588e not known to Sciscape
We reused its verified source archive for OpenNeuro ds003498 version 1.1.1 and its retrieval script. These files supplied event annotations, headers, channel metadata, source URLs, and hashes for a new test of excluded-record count predictions.
- usessci:sha256:74e22fdac1f333c2b0ddb2a290f381a26fb548555b4502c3a28eb34f5d53248c not known to Sciscape
Its test of count variation inside sleep records helped define the prediction question. We reused source selection evidence and electrode scheme text through the later source archive. The present endpoint is prediction of a complete excluded record, rather than count variance inside a record.
- usessci:sha256:a22be20f9664505ea33080cefb407ff7b6665b5b75e6cc0193ba4ef160713581 not known to Sciscape
Its comparison of HFO channel list agreement inside and between records helped motivate exclusion of complete records. We tested count prediction probabilities in the same public source cohort and did not treat this as an independent cohort replication.
- usesdoi:10.18112/openneuro.ds003498.v1.1.1
The analysis used released HFO annotations and channel metadata from OpenNeuro ds003498 version 1.1.1.
- usesdoi:10.6080/K06Q1VD5
The publication identifies CRCNS ieeg-1 as the original data release underlying the OpenNeuro source.
- supportsdoi:10.1038/s41598-017-13064-1
The source study supports the description of subjects 02 through 09 as patients with temporal lobe epilepsy and depth electrodes only.
Cited by
5 publicationsPublic catalog entries on Sciscape.
- usesMean pair coincidence of high HFO mark counts exceeded an independent block shift reference in 18 SEEG sleep records
This work supplied the fixed extra-variance rule used by the direct source publication. We used that rule to define high-count limits from earlier counts.
- usesA fixed variance rule reduced prediction loss for later one-minute counts of unfiltered MNI HFO marks in 18 public SEEG sleep records
This publication supplied the fixed excess variance formula, the Jeffreys count prior, and the cap of 10. We tested that rule with earlier and later bins in a different public dataset.
- usesNominal 80% conformal HFO map bands had 62.5% mean coverage on excluded dates in seven SEEG patients
This study reused the checked source archive, fixed depth-only cohort, count extractor, retrieval code, and training-only excess variance rule. It tested complete count maps on excluded dates, rather than the parent marginal count bands.
- usesExact Gamma product predictions reduced later one-minute MNI HFO count loss in 18 SEEG records under a fixed fit rule
This work supplied the earlier excess variance rule used by our immediate source. We retained its count prior constant, patient variance formula, and cap while changing the marginal probability calculation.
- usesUnder independent patient resampling, the 95 percent interval for mean HFO count prediction gain from a 60 second update included zero in seven SEEG patients
Used the prior whole-record distribution prediction test to define a different question: whether early counts improve later mean predictions within an excluded record. The new test separates total count loss from spatial allocation loss.