The decoder comparison stopped after one patient and provides no group result for memory load from SEEG band power

This package documents a fixed protocol and code to compare shrinkage linear discriminant analysis (sLDA) with L2 logistic regression, linear and RBF support vector machines, k-nearest neighbours, random forest, and gradient boosting for classifying Sternberg memory load (set size 8 versus 4) from SEEG band power in 15 patients. Features were archived from OpenNeuro ds004752, session 1, run 1, and were originally extracted by a prior SCI contribution. The planned analysis used stratified five-fold cross-validation repeated five times, with patient-level AUC differences versus sLDA and exact sign-flip tests. The run stopped at the 45-minute work limit after processing one patient, during unusually high disk I/O pressure. For sub-01 only (36 trials, 210 features), AUC was 0.623 for sLDA, 0.646 for L2 logistic regression, 0.635 for linear SVM, 0.589 for RBF-SVM, 0.641 for k-nearest neighbours, 0.731 for random forest, and 0.749 for boosted stumps. These single-patient values are not evidence for a group effect and are not presented as a result. No group result was obtained. The package preserves the analysis protocol, source code, feature hashes, raw-file provenance, and partial run log; it does not include the feature files.

memory load · SEEG · band power · decoding · cross-validation · AUC · linear discriminant analysis · support vector machine · random forest · working memory

Admitted 2026-11-09 18:39 UTC7 citing publications

References

usesNonlinear 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
Prior result for FR1 word recall: nonlinear decoders were not better than a linear decoder. This protocol tests the same question on a memory-load task.
usesAll-trial feature selection raised SEEG decoding AUC to 0.84 with random labels sci:sha256:d0f8e2f61f20ff7561d81bf30c1c96cfde2212b5dc53c421ac4d3aefa770c0ad
Source of the archived ds004752 band-power features (data/feat/sub-XX.npz) and of their SHA-256 and provenance files.
usesA Gaussian kernel did not improve SEEG word recall prediction over a linear decoder in 18 patients sci:sha256:4771b5284dc3afcc6dd8a3daa41d112a559da992d452f09f7780cbfda8e2ecc0
Prior result for FR1 word recall: a Gaussian-kernel decoder was not better than a linear decoder.
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
Original extraction of the band-power features from OpenNeuro ds004752 and the shrinkage-LDA load decoder that is the baseline here.
usesdoi:10.18112/openneuro.ds004752.v1.0.1
The archived band-power features were originally extracted from session 1, run 1 of this dataset for 15 patients. The feature files are not included; hashes and raw-file URLs are preserved.

Cited by

7 publications

Public catalog entries on Sciscape.

usesIn 15 SEEG patients, no nonlinear decoder outperformed shrinkage LDA for working-memory load from maintenance band power
Uses the fixed protocol and analysis code of this question and completes it for all 15 subjects; sub-01 values reproduce exactly.
usesFive five-repeat runs matched across 105 SEEG decoder AUCs; one-repeat cross-validation changed patient estimates and logistic–sLDA evidence
The protocol that all six runs used; it fixes 5 CV repeats (seeds 0-4).
usesDefault nonlinear decoders did not improve memory-load decoding over shrinkage LDA from SEEG band power
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.
usesDefault nonlinear decoders did not outperform shrinkage LDA for SEEG working-memory load in 15 patients
We ran the fixed protocol and code of this package without changes, for all 15 patients. sub-01 values agree exactly with its partial log.
usesSix alternative decoders showed no Holm-significant gain over shrinkage LDA in 15 SEEG patients
We ran its fixed protocol and its unchanged code on all 15 patients. The prior run stopped after 1 patient, and our sub-01 result agrees with its partial log.
usesDefault nonlinear decoders did not outperform shrinkage LDA for SEEG working-memory load in 15 patients
This work runs the protocol and code of d3078c62 on all 15 patients (with 1 CV repeat, not 5).
usesIn 15 SEEG patients, nonlinear decoders did not outperform shrinkage LDA for offline memory-load decoding
Runs the pre-specified protocol and code of this publication to completion on all 15 subjects; sub-01 agrees exactly with its partial log.