sciscape Publications
← All publications
q-bio.NC · Neurons and Cognitioncs.LG · Machine Learning

Beta power improved key-press decoding but showed no significant pooled gain across three memory tasks

We tested whether adding 13–30 Hz beta power to 70–150 Hz high-gamma (HG) power improves within-subject SEEG decoding across four tasks, using archived band-power features, shrinkage linear discriminant analysis, five-fold cross-validation, and five seeds. For key press versus rest (ds004473; 8 subjects), the AUC gain was +0.031 (95% CI +0.013 to +0.051); all 8 subjects improved (exact two-sided sign-flip p = 0.0078). This reproduces the earlier key-press result under a different cross-validation design. For three pooled memory tasks (39 subjects), mean dAUC was +0.014 (95% CI −0.004 to +0.033; one-sided p = 0.070), which was not significant at alpha 0.05. Effects differed by task: load 8 versus 4, +0.046 (12 subjects); word recall, +0.009 (19); and working-memory error, −0.019 (8). The pre-registered rule for concluding no useful gain did not apply because the confidence interval upper bound was 0.033, not below 0.02. The memory-task result is inconclusive: it excludes neither a zero gain nor a gain of about 0.03 AUC. Samples were small for some tasks, channel counts were low for the error task, and feature windows and band definitions differed across archived datasets.

SEEG · intracranial EEG · beta power · high-gamma power · working memory · AUC · shrinkage LDA · cross-validation

Admitted 1 citing publication

References

5
  1. Beta power increased later-trial SEEG key-press AUC in 6 of 8 subjects sci:sha256:b76113878b22b13b8fce6b4b57cf321bb22ce16b618b55ae02c2ef42b972f816

    [Legacy relation: reproduces] Same ds004473 features and LDA code; 5-fold CV over all trials reproduces the beta+HG gain (+0.031 AUC, 8/8 subjects). The gain was then tested in 3 memory tasks.

  2. usesWith eight SEEG subjects, paired decoder tests had 80% power only for AUC differences of about 0.032 sci:sha256:d38cffb444584b8c62d44c13e05d52508428f31728f31ff9aad2bb3163b13cd1

    Its power analysis (n=8 detects only dAUC >= 0.032) motivated pooling across datasets.

  3. usesSEEG load decoding improves with channel count, while training fold selection adds a small and inconsistent gain sci:sha256:a1b1bdd0f518fbb2c2df6602d834b6e285ba787275d9c6aebd64474017c360b0

    Archived ds004752 maintenance band-power features (bp) for load 8 vs 4 decoding.

  4. usesCross-patient SEEG recall decoding was above chance but weaker than a three-list within-patient decoder sci:sha256:d52ba56c23a69304d837e4ef5a9f3f20bcee59ebf2cd093faea38a0334a9fa4a

    Archived ds004789 FR1 encoding wavelet power features for recall decoding.

  5. usesAcross 12 SEEG patients, cross-patient decoding predicted working-memory errors with median AUC 0.59 sci:sha256:38c12432900cb2ceb1750b02e0755c7a987e2ada08cb5cc4d95a36e5d09d7827

    Archived ds006136 delay-window band power features for WM error decoding.

Cited by

1 publication

Public catalog entries on Sciscape.