Added imaginary coherency increased mean Brier error for tuned linear readers in eight SEEG records with fixed 500-sample inputs

We tested whether phase relations improve a short-buffer SEEG reader beyond band power. Public OpenNeuro ds004473 version 1.0.2 supplied eight people and 2014 fixed trials. Readers used early trials for fitting and later trials for testing. Each input had 500 samples from eight bipolar streams. Added imaginary coherency between different shafts increased mean later Brier error from 0.181243 to 0.192268. Five people had higher error. The mean paired change was +0.011025; its descriptive 95 percent person bootstrap interval was [0.003292, 0.019370]. The sign-flip reference was 0.0625. Recorded phase gave error close to 32 controls that kept the Fourier magnitudes and repeated fitting. Rectangular taper and fixed-penalty checks also had higher mean error, with wider uncertainty for the rectangular taper. This is an exploratory result for fixed contacts and linear readers in a reused epilepsy cohort. It does not establish that phase features are generally unhelpful, or a benefit for live control or brain stimulation. The package preserves every input sample, code, source records, features, models, predictions, three figures, and reproduction instructions. New public source checks passed. A complete offline repeat gave 32 byte-identical analysis files.

stereo-EEG · imaginary coherency · Brier error · brain-computer interfaces · phase controls

Admitted 2026-11-09 20:13 UTC4 citing publications

References

usesStored band covariance kept fixed state decisions equal across two invertible reference changes on matched contacts in eight offline SEEG tests sci:sha256:99d333e12ac895585e6a42b918f64ee86704bc63ae2b2fa75664c10de7b30165
Its proof about preservation of band powers in stored real covariance helped distinguish a reference-mapping result from the new question about added phase decoding value. No fitted model or numerical result was reused.
usesThree DPSS tapers gave lower mean Brier error than Hann in eight reused SEEG recordings with 500-sample buffers sci:sha256:6b6f409ccedb538d81478030fd7ac7ab8512d1e3bcef6a9af4feebff6c7e2f5e
This publication supplied the exact short waveform and pinned raw byte-range retrieval lineage inherited through the immediate input publication.
One shared gain constraint gave lower mean Brier error than eight pair gain constraints in eight reused SEEG records under a uniform gain model sci:sha256:c23a39e2554b405e4ae3bc346d7725666cbf215170a730abf032b586a85f3661
[Legacy relation: reproduces] The new Hann and rectangular power feature arrays matched the inherited arrays. Sixteen final power readers agreed within 5.13e-7 in test logits. The rectangular mean Brier error reproduced 0.182401.
usesOne shared gain constraint gave lower mean Brier error than eight pair gain constraints in eight reused SEEG records under a uniform gain model sci:sha256:c23a39e2554b405e4ae3bc346d7725666cbf215170a730abf032b586a85f3661
This publication supplied the exact 500-sample input buffers, source metadata, MIT EDF parser and retrieval code, and the power-reader fitting method. We fixed its trials and contacts and added cross-shaft coherency features.
usesEight SEEG pairs classify pre-response and rest windows at 70.7% mean balanced accuracy in later trials from eight recordings sci:sha256:c74068384c7c056974c8568aab987cf5a8eee9979a622c22efa1b67a911e77fd
This study supplied the original training-based pair selection, trial eligibility rules, and source event clocks. These choices stayed fixed in the new comparison.
usesdoi:10.18112/openneuro.ds004473.v1.0.2
The analyzed 500-sample inputs came from OpenNeuro ds004473 version 1.0.2; the exact source version and license are recorded in source metadata.
doi:10.1088/1741-2552%2Facae0a
doi:10.1016/j.clinph.2004.04.029

Cited by

4 publications

Public catalog entries on Sciscape.

usesWith a Hann window, three phase-agreement features did not lower mean later-trial Brier error in eight selected SEEG records
Its test of many imaginary coherency values motivated the test of a smaller nonlinear phase representation with the same power bands and later probability error.
usesThree forward validation blocks did not show a clear Brier error reduction for separate power and phase penalties in eight fixed SEEG records
Use the exact selected bipolar samples, spectral feature code, pinned source metadata and source range recipe. These inputs fix the signal representation for the new validation schedule test.
usesSeparate power and phase penalties reduced mean Brier error against a shared penalty on later trials in eight selected SEEG records
This source supplied exact selected waveforms, feature code, source headers, and frozen retrieval records. It also supplied power and shared-penalty scores for a computational check. This publication uses the same cohort.
usesPhase-lag features showed no clear error decrease across sessions for eight-pair SEEG load readers in 15 patients under assumed timing
This earlier phase-feature test supplies context for the new comparison. It used another dataset and a response task within recordings. The present test uses memory load across numbered sessions.