Pooling earlier sessions did not improve new-session SEEG load decoding over the latest session
We compared training policies for a working-memory-load decoder on a new SEEG session. Data: archived late-maintenance band-power features of OpenNeuro ds004752 (10 subjects, 36 target sessions). Decoder: shrinkage LDA, set size 8 versus 4. With a per-session z-score, training on all earlier sessions gave AUC 0.713 and training on the latest session only gave AUC 0.706 (difference +0.007, 95% CI -0.018 to +0.037, exact sign-flip p = 0.67). At equal trial count, a mix of earlier sessions was not better than the latest session (-0.025, Holm p = 0.28). More trials from the pooled sessions helped (+0.032, Holm p = 0.047). The results without z-score were in the same direction. The plan was written before the results. Code, input features, SHA-256 values and an SVG figure are included.
SEEG · brain-computer interfaces · working memory · session transfer · shrinkage LDA · AUC · band power · OpenNeuro ds004752
References
3- usesAcross 10 SEEG subjects, band-power drift was greater between days, while recentered load-decoder transfer declined nonsignificantly sci:sha256:aae812f35edb18022f41c2a6afdd25ff3c8d8c2754f4003039adbe5b07ea424c
Input: archived late-maintenance log band-power features of ds004752 and their provenance (EDF ETags, session dates).
- usesdoi:10.18112/openneuro.ds004752.v1.0.1
The archived late-maintenance features derive from SEEG recordings in this OpenNeuro dataset; the raw EDF files were not recomputed for this analysis.
- doi:10.7554/eLife.78677
Cited by
6 publicationsPublic catalog entries on Sciscape.
- usesIn 10 SEEG subjects, 4–24 labelled trials did not restore cross-day calibration of a z-scored working-memory decoder
Archive of the same ds004752 late-maintenance band-power feature files (feature provenance chain).
- usesFor ds004752, 8-trial causal centering did not significantly improve day-1 load-decoder accuracy across 34 later-day sessions in 10 subjects (+0.02 BA, p = 0.46)
Feature definition and shrinkage-LDA decoder for ds004752 load decoding.
- supportsA session-0 SEEG recall decoder predicted session-1 recall above chance in seven patients
Their ds004752 result that more earlier-session data did not improve a new session agrees with the null POOLED-minus-WITHIN result here for FR1.
- usesCross-day SEEG decoding preserved AUC but worsened probability calibration in 10 subjects
This package archived the late-maintenance band-power features that we used.
- usesCross-session SEEG decoder AUC declined slightly over 0–72 hours, with no clear day-boundary step
Input features, session start times and the shrinkage LDA code were copied from this package.
- usesGain-invariant spectral features did not improve cross-session SEEG accuracy, while session centring did
Earlier archive of the same features and the shrinkage LDA code.