A fixed 20 percent sample mask reduced mean random label balanced accuracy from 73.5 to 50.0 percent in eight SEEG rest recordings with simulated gaps
We tested a fixed sample mask to stop a recording gap rule from entering SEEG signal features. We used eight public recordings from OpenNeuro ds004473 version 1.0.2, with 2,014 retained trials and eight selected bipolar pairs per person. Random labels were independent of source voltage. When those labels set 20 percent sample gaps, mean balanced accuracy on later rest windows was 73.53 percent. Discarding the same sample positions in every window reduced it to 50.04 percent. The paired reduction was 23.49 percentage points, with a descriptive 95 percent person bootstrap interval of 15.42 to 32.94 points. Equal missing counts with different layouts still gave 56.79 percent accuracy; a fixed union gave 49.86 percent. Separate fitted task readers gave 71.94 percent with intact windows and 71.50 percent with the fixed 20 percent mask. The mean task change was minus 0.44 points, with an interval from minus 1.89 to plus 1.02 points. We prove exact gap-rule invariance when all gaps lie inside the fixed mask and required clean samples are unchanged. For unrestricted signals and deterministic decoders without additional information, invariant output must use the common observed support. The package preserves code, integer voltages, provenance, labels, masks, models, scores, three figures, independent checks, and repeat instructions. This is an exploratory offline test with simulated gaps and reused later trials. It does not measure stimulation effects or continuous control.
stereoelectroencephalography · brain-computer interface · signal processing · sample gaps · balanced accuracy · fixed support · random labels
References
8- usesA causal filter gave higher fixed decoder accuracy than fourfold sample removal alone in eight SEEG recordings sci:sha256:e48923d80e223e23c65d73b7a5e6734993070247d942334dabdcfd1ffa26c7b6
This upstream work supplied voltage windows with history and pinned source range records to the direct parent. We used these files through that parent and preserved their retrieval recipe.
- supportsLinear filling gave 74.2 percent balanced accuracy for random labels in eight SEEG rest recordings with 20 percent sample gaps set by label sci:sha256:519436261233c1e63a2fe882315ab0935d972ae4d52615ba8854f7146bd02e3e
A new random label seed repeated above-chance scores when labels set short gaps and below-chance scores when the test rule was reversed. The fixed-mask correction and equal-count layout controls extend the measurement-path evidence on the same cohort.
- usesLinear filling gave 74.2 percent balanced accuracy for random labels in eight SEEG rest recordings with 20 percent sample gaps set by label sci:sha256:519436261233c1e63a2fe882315ab0935d972ae4d52615ba8854f7146bd02e3e
The direct parent supplied exact integer SEEG voltages, pair and trial records, source metadata, fixed task models, range receipts, and numerical helpers. Its random label gap test motivated the new fixed-support correction.
- usesEight SEEG pairs classify pre-response and rest windows at 70.7% mean balanced accuracy in later trials from eight recordings sci:sha256:c74068384c7c056974c8568aab987cf5a8eee9979a622c22efa1b67a911e77fd
This work made the original eight-pair selection, source state windows, trial rules, spectral features, and fixed logistic task models. We kept those inputs and reproduced native features and probabilities before testing the correction.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
This version-pinned OpenNeuro dataset supplied the SEEG source data used in the study.
- doi:10.1088/1741-2552%2Facae0a
- doi:10.3389/fncir.2012.00087
- doi:10.3389/fncir.2013.00006
Cited by
3 publicationsPublic catalog entries on Sciscape.
- supportsShared gaps increased power shift test rejection in independent controls with spectra from 35 SEEG records
The new study adds evidence that common sample support can change BCI feature inference. The prior work tested classifier support and labels. The present study tests power correlation and source independence.
- usesA stationary Gaussian variance correction with 5 percent shrinkage reduced SEEG decoding accuracy for 20 percent simulated sample loss in eight recordings
This direct parent supplied integer voltage windows, selected pairs, trial records, pinned source metadata, range receipts, and fixed logistic decoder states. We removed its 256 history samples. We kept all 500 scored samples and the original decoder.
- usesA stationary Gaussian variance correction with 5 percent shrinkage reduced SEEG decoding accuracy for 20 percent simulated sample loss in eight recordings
This direct parent supplied integer voltage windows, selected pairs, trial records, pinned source metadata, range receipts, and fixed logistic decoder states. We removed its 256 history samples. We kept all 500 scored samples and the original decoder.