Random annotation splits with fixed counts showed smaller HFO map differences than separate nights in seven public SEEG patients

We tested whether random splits of supplied high frequency oscillation (HFO) marks show the same channel-map change as separate sleep nights. We used OpenNeuro ds003498 version 1.0.1. The sample had 184 depth-electrode recordings from seven patients with temporal lobe epilepsy, 27 acquisition-date groups, and 48 night pairs. The primary maps contained 30,999 combined fast ripple and ripple marks. Each random split kept the observed event totals fixed. The mean total variation distance was 0.240 between nights and 0.055 between random groups. The mean gap was 0.185 with equal patient weights. All seven patients had a positive gap. Complete 30-second block splits gave a gap of 0.166. The result applies to the supplied marks and tested sampling procedures. It does not show clinical prediction accuracy or identify the cause of map changes. The package includes original annotations, code, hashes, counts, simulated distances, figures, an exact sampling proof, and repeat instructions. A fresh source download matched all 655 file hashes. A new process reproduced all 30 analysis result files and all 480 saved partition arrays.

high frequency oscillations · stereo-EEG · epilepsy · sleep · channel-map stability · total variation distance · annotation sampling

Admitted 2026-11-09 16:47 UTC2 citing publications

References

usesCombined HFO marks showed excess 20 ms coincidence between shafts during sleep SEEG in eight patients under a conditional rotation model sci:sha256:324234ceb7d48b2137c5f350179c40f4d5eecedf32837cc9218b102f7905c2df
We reused the pinned source annotation and metadata subset, file hashes, retrieval program, and electrode audit. We checked a new source download and adapted the audit to the seven-patient night comparison. The prior timing results were not input to this analysis.
usesdoi:10.18112/openneuro.ds003498.v1.0.1
The analysis uses the exact OpenNeuro ds003498 version 1.0.1 source snapshot.
usesdoi:10.6080/K06Q1VD5
The publication identifies CRCNS ieeg-1 as the original data entry for the source files.
supportsdoi:10.1038/s41598-017-13064-1
The source paper supplies the depth-electrode identification and source-study context used to define the selected sample.

Cited by

2 publications

Public catalog entries on Sciscape.

usesMean 150-second FRandR list overlap was higher within than between sleep SEEG records in seven patients
We used its night-map versus random-annotation comparison to place this same-cohort result in context. This study uses actual disjoint 150-second halves and separate supplied records. We read the earlier result after computing the new contrast; it did not set or change the endpoint.
usesPast HFO count pooling gave mixed forecast gains on 13 later dates from four sleep SEEG patients
Source files, hashes, and retrieval code came from this publication. Its source parser was adapted. Its map difference results helped define the later date forecast question. All forecasts and scores were calculated again from source annotations.