Training on late grasp windows raised mean late test balanced accuracy from 53% to 64% with eight SEEG inputs in four people
We tested task training time in public SEEG recordings from four people. Both models used the same eight inputs, power bands, training blocks, sample count, and logistic regression settings. Early training used 0.5 to 1.5 s after the target start. Late training used 2.5 to 3.5 s. Mean late test balanced accuracy rose from 52.8% to 63.8%. All four person means improved. One of eight grasp tests became worse. Mean early test balanced accuracy fell from 79.2% to 73.7%. An intercept adjustment gave 58.3% late balanced accuracy. A reader of measured force gave 98.1% late balanced accuracy. The exact person sign-flip reference was p=0.125. Four people give weak population evidence. The parent study already used the test blocks. This follow-up is not an independent confirmation. The public inputs have no contact map. The result supports tests at both task times. It does not establish live control or the source of the useful signal. The package contains exact inputs, source receipts, code, models, all scores, three figures, checks, and repeat instructions.
stereoelectroencephalography · grasp detection · brain-computer interface · training time · balanced accuracy · spectral power · offline classification
References
5- usesFixed SEEG readers scored 80% mean balanced accuracy early and 54% later across executed grasps in four people sci:sha256:9317803e3e59d0db5ce4f76b02220e1344da603fb253dca80e67acfbce823766
This study supplied the exact compact features, raw audit files, source metadata, extraction code, and block split. Its early and late test scores led to this question. We fit late task readers with the same inputs and score the same test blocks. The early balanced accuracy and all eight input sets match the parent.
- usesRest calibration reduces mean task accuracy in fixed SEEG imagery tests from two people sci:sha256:b2d5a1ce684c63615eb48aace1fd2a51169cfeb20f8e08e74a6759aa6964ec06
The direct parent inherited the extraction method and A/B features from this study. This package retains that code and data lineage and its license terms.
- usesdoi:10.82901/nemar.nm000397
The analysis used NEMAR dataset nm000397, version 1.0.0; the source record gives this version-specific dataset DOI and its CC0-1.0 license.
- usesdoi:10.5061/dryad.nq4fs
The NEMAR source record identifies this Dryad DOI as the original data source for the analyzed conversion.
- usesdoi:10.1371/journal.pone.0150359
This paper describes the source SEEG grasp recordings and task used by the study.
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesMixed training reduced the larger early or late test Brier error from 0.325 to 0.255 with equal window counts in SEEG data from four people
This package supplied the compact features, raw audit windows, source manifest, metadata, extraction code, and previous models. We repeated its early and late-selection readers. The new mixed readers used one assigned task window per trial throughout selection, scaling, and fitting. The test blocks were already used by this parent.