Negative binomial forecasts reduced mean HFO count loss on later dates at contacts 1–2 in seven SEEG patients under a fixed history model
We tested probability forecasts for released high frequency oscillation (HFO) overlap marks. We used contacts 1–2 on each depth electrode shaft in OpenNeuro ds003498 version 1.1.1. Seven patients supplied 37 fit records and 98 test records. Fit records came from the first released date. Test records came from later dates. Both models used the same predicted mean from 30 seconds of past counts. The negative binomial model reduced mean 10-second test loss from 0.768525 to 0.753555 nats. The gain was 0.014970 nats. The exact empirical patient bootstrap interval was 0.005893 to 0.024756 nats. Six patients improved; one fit selected Poisson. Nominal 90 percent interval coverage was 96.93 percent for Poisson and 97.56 percent for negative binomial. The result supports added count uncertainty under the fixed model. It does not establish correct clinical calibration or a biological cause. The package preserves code, locked source inputs, all forecasts and checks. Complete repeats from archived inputs and new public downloads gave identical analysis files.
high frequency oscillations · SEEG · negative binomial · Poisson model · count forecasting · probability forecast · dispersion · epilepsy · electrophysiology
References
6- usessci:sha256:74e22fdac1f333c2b0ddb2a290f381a26fb548555b4502c3a28eb34f5d53248c not known to Sciscape
The earlier source and cohort audit supplied implant evidence and source exposure checks through the prior forecast package. We preserved its source documents, checked the hashes and downloaded the public inputs again.
- usesPast counts on other shafts did not reduce mean 10-second HFO overlap forecast loss in seven SEEG patients under a fixed Poisson model sci:sha256:12a7d289dc68acda3f1a21524bc6c5f50a0ed21acd61b932b847907da35ce3c2
This publication supplied the locked ds003498 source package and the fixed recent-history mean settings. We rebuilt separate-pair counts from source events. We added a first-date split, a fixed-mean negative binomial probability test and calibration checks. We did not use prior derived predictions as outcomes.
- usesdoi:10.18112/openneuro.ds003498.v1.1.1
The analysis uses the released event marks and metadata from OpenNeuro ds003498 version 1.1.1.
- usesdoi:10.6080/K06Q1VD5
This is the original CRCNS ieeg-1 dataset identified by the source package for the released data.
- usesdoi:10.1038/s41598-017-13064-1
This paper describes the source HFO event marks and their relationship to surgical outcome; surgical outcome was not an analysis target here.
- doi:10.1088/1741-2552%2Fac520f
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesAdded calibrated ripple marks reduced mean HFO map log loss by 0.018 nats per mark on later dates in seven SEEG patients
Its first-date fit and later-date test informed the date split. Our target is a conditional channel map on three mesial pairs per shaft. We do not reproduce its short-bin count-burden forecast.
- usesCopulas reduced mean HFO forecast loss on later dates at contacts 1–2 in seven SEEG patients with matched shaft probabilities
We used its pinned source archive and unchanged helper code. The helpers fit marginal models and calculate uncertainty. We rebuilt source counts and computed new joint forecasts. Prior derived predictions were not observations.