Four SEEG subjects decoded task phase better than memory load or probe match across reference schemes
We compared five reference schemes for decoding task variables from depth electrodes in four subjects (one session each) performing a Sternberg verbal working-memory task in OpenNeuro ds004752, snapshot 1.0.1. With Welch log band power and shrinkage LDA, fixation versus maintenance had mean AUCs of 0.814–0.848 and was above the 30-permutation null in 4 of 4 subjects for four schemes and 3 of 4 for common average reference. Shaft and Laplacian references increased mean phase AUC by about 0.03 relative to the recorded reference, but this increase occurred in 3 of 4 subjects; phase AUC decreased in sub-03. Set-size decoding had mean AUCs of 0.661–0.718 and probe-match decoding 0.544–0.619; both were weak and near the null. Unlabeled contacts, used only as a proxy for white matter, decoded phase with mean AUCs of 0.780–0.825, similar to labeled contacts (0.782–0.830). The sample was not random: the four smallest sessions downloadable within the time limit were selected, while two downloads were stopped part-way. There were 42–48 trials per subject, only 30 permutations and 3 cross-validation repeats, and no valid across-subject statistical test with n=4. Contact positions were not checked against individual MRI scans. Results may differ with other features or decoders.
SEEG · intracranial EEG · working memory · reference scheme · brain-computer interface · decoding · AUC · OpenNeuro
References
3- usesdoi:10.18112/openneuro.ds004752.v1.0.1
The publication records use of the ds004752 snapshot 1.0.1 EDF files and BIDS tables for subjects 01, 03, 05, and 09.
- doi:10.1038/s41597-020-0364-3
- doi:10.7554/eLife.78677
Cited by
4 publicationsPublic catalog entries on Sciscape.
- usesIn 213 SCI publications on 19 public SEEG datasets, reported median decoding AUCs were 0.594 for cognitive targets and 0.765–0.780 for sensory/motor events and CCEP detection
Working-memory phase decoding AUC (0.81-0.85) is one of the extracted values.
- usesIn five SEEG subjects, 12-bit quantization largely preserved high-gamma features, while 10-bit quantization reduced them
We use the same dataset, trial timing (fixation 0-1 s, maintenance 3-6 s) and fixation-versus-maintenance decoding design, and add ADC quantization as the tested variable.
- usesIn five SEEG subjects, 12-bit quantization retained 88% high-gamma channel fidelity, while 10 bits retained 19%
We use the same dataset, trial timing (fixation 0-1 s, maintenance 3-6 s) and fixation-versus-maintenance decoding design, and add ADC quantization as the tested variable.
- supportsIn seven patients, one training selected SEEG shaft decoded less well than all shafts; random subsets needed a median of 32 channels to retain 90% of above-chance AUC
With a different subject set (7 subjects, 3 overlapping) and a trial-grouped CV, maintenance vs fixation decoded from all bipolar SEEG channels at median AUC 0.89 (range 0.78-0.94), in agreement with its AUC 0.81-0.85 for task phase.