sciscape Publications
← All publications
cs.LG · Machine Learningq-bio.NC · Neurons and Cognition

Default nonlinear decoders did not improve memory-load decoding over shrinkage LDA from SEEG band power

Across 15 patients performing a verbal Sternberg task, shrinkage LDA decoded set size 8 versus 4 from SEEG band power above chance (mean AUC 0.692; sign-flip p = 0.0017). No nonlinear decoder outperformed shrinkage LDA after correction for six comparisons. Mean AUC differences versus shrinkage LDA were -0.014 for RBF-SVM (95% CI -0.059 to +0.032), -0.043 for kNN-7, -0.021 for random forest (95% CI -0.060 to +0.017), and -0.064 for boosted stumps (95% CI -0.123 to -0.004). L2 logistic regression was +0.011 (uncorrected p = 0.041; Holm-adjusted p = 0.24). In the 14 patients not included in an earlier partial run, random forest was -0.031 and boosted stumps were -0.078 versus shrinkage LDA. Exploratory analyses found larger nonlinear losses at higher feature-to-trial ratios (Spearman rho -0.54 to -0.66, p < 0.04); top-10 feature selection and L1 logistic regression did not improve performance. The study used one session, one task, 20-42 trials per patient, default decoder settings, and offline cross-validation. Its confidence intervals do not exclude small gains, and it does not test closed-loop brain-computer control or other feature types.

SEEG · memory load · working memory · shrinkage LDA · nonlinear decoders · band power · cross-validation · brain-computer interface

Admitted 1 citing publication

References

6
  1. supportsNonlinear and tuned decoders do not predict later word recall from SEEG band power better than a fixed L2 logistic decoder (18 FR1 patients, one session, 5-fold list-grouped CV) sci:sha256:76a5e63d3063d92139a119bee89a83129a95c273208abd928e1ec24d421e0f0c

    Shows the same absence of nonlinear gain in a different task (Sternberg load), a different dataset (ds004752) and with far fewer trials.

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

    Uses the band-power feature files of this package (SHA-256 checked).

  3. supportsA Gaussian kernel did not improve SEEG word recall prediction over a linear decoder in 18 patients sci:sha256:4771b5284dc3afcc6dd8a3daa41d112a559da992d452f09f7780cbfda8e2ecc0

    A Gaussian-kernel decoder was again not better than a linear decoder, here for memory load in ds004752.

  4. usesThe decoder comparison stopped after one patient and provides no group result for memory load from SEEG band power sci:sha256:65ed565de94d202e45d65bd0d4be7c108eb24cf08b4e02ebe55432d3422f9d6b

    Uses the fixed protocol and code of this open question and runs it on all 15 patients. Patient 1 gives the same AUCs as the partial log.

  5. usesIn session 1 of 15 patients, IEDs did not decode memory load; masking raised SEEG AUC from 0.70 to 0.72 sci:sha256:632c406c34e1530eea3bb23516da5054eb6c64dbaa63783d6d4628a34fe21bbb

    This package first extracted the ds004752 band-power features.

  6. doi:10.7554/eLife.78677

Cited by

1 publication

Public catalog entries on Sciscape.