Pair models with a uniform timing benchmark lowered overlap label loss on later dates in five of seven SEEG patients
This study tested whether electrode pairs retain fast ripple overlap information across recorded dates. It used public event marks from OpenNeuro ds003498 version 1.0.1. The primary cohort had seven patients, 184 sleep runs, and 59,714 test fast ripples. The cohort follows paper implant records; BIDS metadata labels the channels ECoG. Both models used an exact chance estimate for a uniform shift of fast ripple time. A model with one fitted value per pair had lower test log loss in five of seven patients. Mean gain was 0.06234 nats per fast ripple. The primary patient sign-flip test gave p = 0.109375 under patient independence and symmetric null gains. The evidence for a population gain is limited. Patient 01 was excluded after source review showed a mixed implant. The package preserves the first analysis and the correction record. It also preserves source tables, hashes, new code, predictions, figures, and reproduction instructions. A fresh source retrieval matched 655 files. A second process gave 14 identical numerical files. This label study does not test neural causation or clinical prediction.
fast ripples · intracranial EEG · electrode pairs · event overlap · log loss · epilepsy · reproducible research
References
3- usesDispersed 30-second samples had lower FRandR rate error in five of nine sleep SEEG patients sci:sha256:5f37b3d8f7114cd339fa02ff6c40c1b53aaa0363b131a423d868aaf944892808
This prior work supplied the initial ds003498 source tables, pinned manifest, and MIT retrieval script. We adapted the script and checked a fresh source download. The new study tests overlap label loss across dates in a corrected depth electrode cohort. It reuses patients and is not an independent patient replication.
- usesdoi:10.18112/openneuro.ds003498.v1.0.1
Source event tables, recording metadata, acquisition dates, contact status, and the voltage inspection example were taken from OpenNeuro ds003498 version 1.0.1. The saved source manifest pins the Git commit and file hashes.
- supportsdoi:10.1038/s41598-017-13064-1
This paper supplies the implant classification used to define the depth electrode cohort and the source FRandR label context.
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesEqual recording weights reduced FRandR map loss on later dates in three of seven sleep SEEG patients with a fixed 1 percent uniform mixture
This publication supplied the earlier-half calendar-date split and record-duration interpretation. We applied those choices to spatial FRandR channel shares. Its overlap-label models and scores were not used as our prediction models or results.
- usesFast ripple duration lowered loss for supplied overlap labels on later dates in seven SEEG patients
Used the preserved source event tables and metadata, retrieval code and manifest, implant classification, source parsing and circular timing definitions, and patient date split design. Added individual-duration models and independent numerical checks in the same seven-patient cohort. This is not an independent patient replication.