Selection of eight SEEG pairs before four-fold validation increased scores for shuffled load labels in 15 patients
Selection of eight SEEG pairs before four-fold validation gave higher scores for random load labels than selection inside training folds. We tested a fixed three-band logistic model in 15 patients from OpenNeuro ds004752 version 1.0.1. There were 200 independent label draws per patient. The mean balanced accuracy was 64.3 percent for selection before validation and 50.2 percent for selection within training folds. The paired difference was 14.1 percentage points, with a patient bootstrap 95 percent interval of 12.2 to 16.3 points. We prove an exact expected score of 50 percent for the training-only conditional label control. With recorded labels, valid fold selection gave 53.3 percent and a separate-session test gave 52.8 percent. These scores do not establish a stable memory-load readout. The result extends prior selection-bias evidence to a separate SEEG cohort. The package preserves data features, fixed folds, labels, predictions, fitted states, source hashes, raw retrieval tools, and figures. The full repeat matched all results. Fresh raw checks cover one trial in each of 30 sessions. Signal phase timing is assumed. The study tests offline readout evaluation and does not test stimulation.
SEEG · stereoelectroencephalography · contact selection · cross-validation · selection bias · balanced accuracy · brain-computer interface · verbal working memory
References
6- Contact selection before cross validation raised estimated balanced accuracy by 15.6 percentage points in eight SEEG recordings with 32 calibration trials sci:sha256:2a94c909e0019de38093ca2e8167cc4b87ac1b4d7532223ee7271179ec4267f0
[Legacy relation: reproduces] We repeated the comparison of eight-pair selection before validation with selection inside training folds in a separate cohort and verbal working-memory task. We used unequal class counts and single-trial labels. This is a conceptual replication, not a repeat of the original numerical result.
- usesEight SEEG pairs showed no clear gain over fixation for letter load across sessions in 15 patients under assumed epoch timing sci:sha256:1c0c74901b6082055c1889d816c6520f050a9fff799c8dce26e8f044f7ce2e9a
We used its verified 30-session power feature files, source metadata, raw interval receipts, and extraction tools for a new contact-selection test. We kept its first and last numbered sessions. We used a different trial filter and three power bands.
- usesTest labels in pair selection raised mean random-label SEEG accuracy in eight recordings with 32 calibration trials sci:sha256:00ce70d14ab1e4dce86b1a5ad0d232bda08db998c876b2dda79b1eac01530b2a
Its random-label control and training-state checks informed our label control and held-out-label mutation audit. We added an exact chance proof for independent permutations within fixed folds.
- usesdoi:10.18112/openneuro.ds004752.v1.0.1
The study uses the first and last numbered sessions from this exact dataset release.
- doi:10.1038/nn.2303
- doi:10.7554/eLife.78677
Cited by
3 publicationsPublic catalog entries on Sciscape.
- usesA 12-trial confidence update lowered Brier error versus an offset but had higher error than a class-rate constant in 15 SEEG session tests at C = 0.1 under assumed timing
This work is the source of the inherited training-based contact selection. Our results assume those fixed lists. We did not repeat or optimize contact selection.
- usesLog band contrasts before averaging canceled separate positive scalar window gains; Brier error did not clearly decrease versus pooling in 15 SEEG tests at C = 0.1 under assumed timing
The fixed eight bipolar pair choices and the clean, previous-correct trial rule follow this training-only selection analysis. The current work preserves these choices and does not select pairs again.
- usesPhase-lag features showed no clear error decrease across sessions for eight-pair SEEG load readers in 15 patients under assumed timing
This study supplies the prior-correct trial rule and the three-band method for pair selection. We apply the rule to the first and last sessions. We select eight pairs from the fitting session only.