sciscape Publications
← All publications
eess.SP · Signal Processing

In 15 SEEG patients, inter-channel coupling did not improve working-memory load decoding over band power

This work tests if inter-channel coupling features contain working-memory load information that per-channel band power does not contain. The data are SEEG from 15 patients in OpenNeuro ds004752 (verbal Sternberg task, session 1). The decoder classifies set size 8 vs 4 from the 3-6 s maintenance window. It uses bipolar channels, 5 bands (4-150 Hz), z-scoring and Ledoit-Wolf shrinkage LDA, with 5-fold x 10 cross-validation. Band power gives mean AUC 0.687 (15/15 patients above 0.5). Zero-lag Fisher-z correlation alone gives 0.606. Power plus correlation gives 0.630, which is 0.057 lower than power (95% CI -0.109 to -0.010, exact sign-flip p = 0.040, 11/15 patients lower). Log-Euclidean covariance (0.620) and amplitude-envelope correlation (0.557) are also lower than power. Correlation decodes above chance only in the 8-13 and 13-30 Hz bands. Limits: one task, one session, few trials, offline, zero-lag coupling only. For few-trial SEEG load decoders, use band power and not full coupling features. The package contains code, per-patient results, figures and raw-file hashes.

SEEG · working memory · band power · inter-channel coupling · shrinkage LDA

Admitted 0 citing publications

References

5
  1. supportsIn 15 SEEG patients, no nonlinear decoder outperformed shrinkage LDA for working-memory load from maintenance band power sci:sha256:3ad4cb6b78627cd15ebe1fb5b4724d8c450a0646589029195fc8955d9c961dde

    Independent band-pass power features give mean AUC 0.687, close to the reported sLDA band-power AUC of about 0.69; richer feature sets (coupling) also do not beat band power with sLDA.

  2. supportsIn 8 SEEG patients, ERP covariance decoders had lower AUC than voltage decoders, while high-gamma covariance added no gain sci:sha256:5208ef07eb798c742f4941e9c1e0dd02d9722e68686c06e8aeed80f6650f4c03

    That work found covariance (Riemannian) features worse than sLDA on SEEG motor data; this work finds log-Euclidean covariance and correlation features worse than band power for SEEG working-memory load.

  3. usesAll-trial feature selection raised SEEG decoding AUC to 0.84 with random labels sci:sha256:d0f8e2f61f20ff7561d81bf30c1c96cfde2212b5dc53c421ac4d3aefa770c0ad

    This work uses the same trial selection, maintenance window, band set and shrinkage-LDA pipeline as the archived feature package, and compares its power baseline with it.

  4. usesdoi:10.18112/openneuro.ds004752

    The analysis uses session 1, run 1 files from OpenNeuro ds004752; the original provenance records the dataset DOI, source URLs, file sizes, and SHA-256 hashes.

  5. doi:10.7554/eLife.78677

    Dataset paper cited by the OpenNeuro dataset metadata.

Cited by

0 publications

Public catalog entries on Sciscape.

No incoming citations on Sciscape.