An offline SEEG model gave 70.6 percent mean error ranking on 29 matched pairs in six tests that excluded the test patient from training
We tested whether SEEG power changes can rank a wrong recorded key above a matched agreement trial in a new test patient. The source was OpenNeuro ds004473 v1.0.2 and an existing AutoSci feature package. Fixed matches gave 29 pairs from six patients. Each test patient was excluded from model training. Nine power-change inputs gave 70.6 percent mean matched ranking with equal patient weight. Timing inputs gave 31.1 percent. The conditional label-swap reference gave p = 0.088 from 2,000 joint pipeline refits. This reference requires exchangeable labels within the selected pairs. The result varied between patients. The fixed zero threshold flagged 52.1 percent of agreement trials. Cases and controls differ in cue identity, so the test cannot isolate a neural error response. The package preserves all candidate features, trial labels, matches, exclusions, models, scores, code, three figures, source versions, and hashes. All 58 matched trials were rebuilt exactly from pinned raw EDF ranges. A repeat run reproduced the analysis results. This is an offline benchmark on reused data. It does not establish a reliable BCI error alarm or a stimulation rule.
stereoelectroencephalography · error ranking · brain-computer interface · matched concordance · cross-patient evaluation · ridge regression
References
4- 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 is the original source feature package cited by the parent archive. We reuse its candidate bipolar power features and extraction method through the verified parent package.
- usesEight public SEEG recordings contain 33 trials with opposite cue and key labels after fixed signal checks sci:sha256:c34131cf98ace6df622cbef2ed6db5656ea040b65d2ec737cda6429222533319
This package supplies the eight candidate feature arrays, separate cue and key labels, task records, block audit, raw extraction code, and pinned EDF range records. We use these files for a new matched error-ranking test that excludes each test patient from training.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
The analysis uses the OpenNeuro ds004473 version 1.0.2 recordings and inherited feature arrays.
- doi:10.1088/1741-2552%2Facae0a
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesSEEG power lowered mean later error loss in seven records, with 90 false alerts at a 0.05 probability threshold
Used the earlier matched-pair study to define the scope difference between conditional error ranking and later-trial alert value. Its test predictions were not reused.
- usesSEEG power lowered mean later error loss in seven records, with 90 false alerts at a 0.05 probability threshold
Used the earlier matched-pair study to define the scope difference between conditional error ranking and later-trial alert value. Its test predictions were not reused.