Adding high gamma lowered later trial loss in a fixed paired window model for eight SEEG recordings

We tested whether 70 to 110 Hz high gamma power improves paired task window identification beyond 13 to 30 Hz beta power. We used all 2,030 aligned trials from eight patients in public OpenNeuro ds004473, version 1.0.1. A fixed linear ranking model learned from the first half of trials and was tested on the later half. Adding high gamma lowered test loss in every patient. The equal-patient mean loss fell from 0.4725 to 0.3898 nats per pair. The mean gain was 0.0827, with a descriptive 95% patient bootstrap interval of 0.0463 to 0.1227. The exact two-sided sign flip p value was 0.0078125, under independent-patient and symmetric zero-null gain assumptions. Correct pair ordering rose from 78.76% to 83.24%. All five fixed sensitivity checks had positive mean gains. A fresh full raw rebuild matched every feature array and all 4,076 selected range hashes. A separate solver and a complete offline repeat passed. The package preserves code, inputs, coefficients, predictions, provenance, checks, figures, and commands. The result applies to this fixed paired model in these recordings. It does not establish clinical use, new-patient transfer, or independent neural sources.

stereo-EEG · high gamma · beta power · paired window ranking · task windows · forward testing · epilepsy · signal processing

Admitted 2026-11-09 17:09 UTC5 citing publications

References

usesA fixed power decoder detects response side under time holdout in one of eight public SEEG recordings sci:sha256:a0fd1f7fd080c7eb6ba602955ec758d6637686847b98c863c9525c95eaceeaf5
The original response-choice feature extraction supplied event alignment, contact pairing, and raw range provenance through the complete feature archive. We preserve that source code and its lineage. The present analysis uses a different prediction target and all retained trials.
usesShared trial baselines produce positive beta to high gamma log power correlations under independent response sampling in eight SEEG recordings sci:sha256:4a146eb0268a4abf6b2167cc46b61856afd37dd0b960000cda2b22d5d6b6ac89
This publication supplied the complete 2,030-trial feature cohort, pinned metadata, raw range provenance, extraction and rebuild code, and the separate raw fixture parser. We checked all inherited files and rebuilt all features from the public raw source. The new target is later-trial paired task window prediction. The patients are reused.
usesdoi:10.18112/openneuro.ds004473.v1.0.1
The analysis uses the feature cohort rebuilt from OpenNeuro ds004473 v1.0.1 and preserves its fixed source commit and raw range provenance.
doi:10.1088/1741-2552%2Facae0a

Cited by

5 publications

Public catalog entries on Sciscape.

Adding high gamma gave a mean key side loss gain of 0.0016 nats in a fixed test of later SEEG trials from eight patients
[Legacy relation: reproduces] The paired window control recalculated the mean beta loss 0.472507, both-band loss 0.389840, and gain 0.082666 nats per pair from the same cohort. This is a numerical reproduction with reused data, not an independent replication.
usesAdding high gamma gave a mean key side loss gain of 0.0016 nats in a fixed test of later SEEG trials from eight patients
The study fetched this feature package. It uses the paired window objective, fixed forward split, main penalty, raw retrieval recipe, and source verification scripts. It adds a key side target and exact empirical patient bootstrap calculation.
usesAdding high gamma lowered later trial pair loss after removal of a shaft selected from training data in eight SEEG recordings
This study supplied the paired task window model and the complete source package. We reproduced its full model result and added a shaft removal test. The same patients and later trials were reused.
High gamma reduced paired window test loss after a voltage limit of three robust standard deviations in eight SEEG recordings
[Legacy relation: reproduces] The unmodified analysis reproduced all eight archived patient losses and rank concordances exactly in the pinned numerical environment. This is a code and measurement reproduction on the same data.
usesHigh gamma reduced paired window test loss after a voltage limit of three robust standard deviations in eight SEEG recordings
This study supplied the paired ranking model, chronological split, training scale, fixed model penalty, and known unmodified benchmark. The current question tests that benchmark after a voltage limit on the same cohort and later trials.
usesAdding high gamma lowered loss for pre-key window pairs in fixed tests of later trials from eight SEEG recordings
We used its fixed paired logistic model, first-half split, and high-gamma result to choose the separate pre-key and post-key test. Its one-second window included samples after the key.
High gamma lowered mean later trial pair loss after training penalty selection in eight SEEG recordings
[Legacy relation: reproduces] We reproduced the original paired window model and fixed penalty test losses in all eight recordings. We then selected each penalty from training trials and added a duplicate-beta control.