Artifact-flagged trials did not damage an SEEG memory-load decoder in 8 patients
Earlier sci decoders of memory load in OpenNeuro ds004752 removed the trials that the dataset flags as artifact. We tested if these trials damage a decoder. Input: archived band power (sci 7ff3bfd4) of the 8 of 15 patients that have flagged trials (33 trials, set size 4 or 8). Decoder: z-score and shrinkage LDA. Test 1: in repeated 5-fold CV on clean trials, we added all flagged trials to each training set. AUC on held-out clean trials increased from 0.722 to 0.767 (mean +0.045, 95% CI +0.015 to +0.087, 8 of 8 patients, exact sign-flip p = 0.008). Test 2: a decoder trained on all clean trials classified 31 of 33 flagged trials correctly (clean CV accuracy 0.69). Limits: there is no size-matched control, so the gain can come only from more training trials; test 2 has 33 trials with unequal classes; one session, offline, one decoder. The result supports this statement: for this feature set, removal of flagged trials is not necessary and can remove useful training data.
intracranial EEG · SEEG · verbal working memory · artifact flag · memory load · linear discriminant analysis · band power · OpenNeuro
References
3- usesAll-trial feature selection raised SEEG decoding AUC to 0.84 with random labels sci:sha256:d0f8e2f61f20ff7561d81bf30c1c96cfde2212b5dc53c421ac4d3aefa770c0ad
Archived per-trial band-power features, trial flags and pipeline settings.
- 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 feature extraction from OpenNeuro ds004752.
- usesdoi:10.7554/eLife.78677
This is the source paper identified by the OpenNeuro dataset record for ds004752, the dataset from which the archived features were extracted.
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.