All available SEEG pairs had an uncertain mean task-state accuracy gain over eight pairs after 32 calibration trials with L2 C=1 in eight records
We tested a choice for brain-computer interface design: more signal inputs after short calibration. We used eight clinical SEEG records from OpenNeuro ds004473 v1.0.2. We compared eight selected bipolar pairs with all 100–112 available pairs. Models used only the first 32 retained trials for selection, scaling, and fitting. Both models used a fixed L2 penalty with C = 1. The same 674 later trials tested every model. Mean balanced accuracy was 62.6% with eight pairs and 64.2% with all pairs. The paired mean change was +1.6 percentage points. The 95% person-bootstrap interval was -0.1 to +3.4 points. This interval includes zero. We also tested 1–64 pairs, 16–64 calibration trials, and other penalty choices. When calibration selected the penalty, both input models reached about 64.7% mean accuracy. These checks are descriptive. The evidence concerns event-based task states in this cohort. It does not establish live movement-intent detection. The package preserves features, raw audit ranges, source records, code, models, predictions, checks, and figures.
stereoelectroencephalography · brain-computer interfaces · input count · calibration · balanced accuracy · logistic regression · task-state decoding
References
4- usesEight fixed SEEG decoders scored 65.8% mean balanced accuracy for response versus rest windows after task condition changes sci:sha256:27c3a162d5f23c03418feb1337bb50c3dbd6fcf78da3fdc3bdc12da973491084
This earlier task-condition transfer test guided the interpretation of early calibration and later test conditions in the same cohort. We used its task-state scope. We did not reuse its fitted models or test split.
- usesEight SEEG pairs classify pre-response and rest windows at 70.7% mean balanced accuracy in later trials from eight recordings sci:sha256:c74068384c7c056974c8568aab987cf5a8eee9979a622c22efa1b67a911e77fd
We reused the exact public power features, trial records, source metadata, raw audit ranges, and retrieval code. We made new model fits, input-count comparisons, short calibration splits, penalty checks, and figures.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
Public SEEG task dataset analyzed at version 1.0.2; the source metadata records this exact version and CC0 license.
- doi:10.1088/1741-2552%2Facae0a
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesAdded fitting trials reduced Brier error more than added selection trials in eight offline SEEG tests with eight pairs and a fixed penalty
This report informed the scope of training-only pair selection. We separate the eight selected pairs from the full candidate set needed for calibration.
- Trial ensembles reduce SEEG probability error with eight fixed pairs after 32 calibration trials in eight recordings
[Legacy relation: reproduces] The single eight-pair and all-pair readers repeat the C=1 first-32-trial comparison on the same final-third test trials. Their equal-patient accuracy values match the prior 62.6% and 64.2% results. They serve as checks and baselines for the new probability and input-budget comparisons.