A fixed offline SEEG decoder reached 53.8 percent balanced accuracy before key responses in eight participants with task blocks held out
We tested a fixed offline decoder of left and right key labels in public SEEG data. The test used all eight participants and 2,030 trials in OpenNeuro ds004473 version 1.0.2. The inputs were bipolar spectral power from 350 to 50 ms before the stored response. Complete task blocks were held out. Mean balanced accuracy was 53.84 percent. The participant bootstrap 95 percent interval was 50.36 to 58.44 percent. The conditional label permutation p-value was 0.005 with 199 permutations. The before-cue control reached 51.94 percent. The paired gain was 1.89 percentage points, with an interval from -3.49 to 8.10 points. A post hoc check without participant 2 gave 51.85 percent and p = 0.130. A generic correct-flag filter would remove all left-key trials for participant 1. The result supports a small fixed benchmark. It does not establish online control, independent motor intention, or transfer to new people. The package preserves three figures, features, predictions, code, source versions, range hashes, audits, and reproduction commands.
stereoelectroencephalography · brain-computer interface · key decoding · balanced accuracy · block-held-out validation · permutation test
References
2- usesdoi:10.18112/openneuro.ds004473.v1.0.2
The analysis used behavior records, events, channel metadata, and version-pinned SEEG signal ranges from all eight participants in OpenNeuro ds004473 version 1.0.2.
- doi:10.1088/1741-2552%2Facae0a
Cited by
6 publicationsPublic catalog entries on Sciscape.
- usesA fixed pre-cue SEEG power method did not establish lower response-time error than task history in eight final blocks
This work supplies the public behavior metadata and checked block-size rule. They enable causal task-history inputs and the final-block split.
- usesFixed SEEG decoders had higher mean early cue call rates than rest call rates on 494 matched later trials in eight people
We used its source metadata, retained trial tables, contact pair tables, trial exclusion audit, and fixed source object versions. We adapted its EDF reader to extract new 200-sample windows. We did not use its features or decoder scores.
- usesWith a shared penalty, added SEEG power increased mean loss for 0.5-second response forecasts in eight cued task recordings
This publication supplied public behavior logs. We used them to find response entries without a registered key press and to define the exploratory corrected target.
- usesRolling error bounds did not establish a lower mean interval score for response time in eight final SEEG task blocks
This work supplied the behavior files and checked block rules through the direct input publication. They enabled the final task block split and the reconstruction of past response history.
- usesAdding SEEG power through 200 ms increased squared log response time error by a mean 0.69% on later trials in eight recordings
Its key-response timing and correctness-field checks informed our timing audit and our choice to retain trials without selection on current correctness.
- usesEight selected SEEG channels gave 56.2% mean balanced accuracy for cue direction on later trials in eight sessions
We read this study after the main decoder test. Its label and timing audit prompted an extra check for zero Response-minus-Go-Cue intervals. We did not reuse its code, features, or decoder results.