With eight pairs, omitting 70–110 Hz power reduced mean balanced accuracy by 5.6 points in eight offline SEEG tests
We tested public SEEG data from eight patients in OpenNeuro ds004473 version 1.0.2. Each decoder selected eight bipolar pairs from early training trials. All models used the same later test trials. Power at 8–12, 13–30, and 70–110 Hz gave 70.7% mean balanced accuracy. Omission of 70–110 Hz gave 65.1%. The paired mean change was −5.6 percentage points, with a descriptive 95% patient bootstrap interval of −8.5 to −2.6 points. The exact two-sided patient sign-flip p value was 0.0234. Accuracy fell in seven of eight patients. Fixed-pair and fixed training-budget tests are also preserved. The package includes complete features, source versions and hashes, raw audit samples, code, models, predictions, three figures, validation results, and reproduction instructions. The evidence concerns offline task states in quality-selected trials. It does not establish online intent detection, hardware energy savings, or a stimulation result.
stereoelectroencephalography · brain-computer interfaces · gamma power · balanced accuracy · logistic regression · offline decoding
References
7- usesA rest threshold set for 5 percent errors gave 3.9 percent false detections and 26.5 percent sensitivity in eight later-trial SEEG tests sci:sha256:2100f971aed9c12a7d4be6af5c0b8c155e730ec5176d7b8c6dd80fcfd083230d
We reused the complete 2,014-trial feature set, extraction code, source records, and raw audit samples. We verified all 118 imported feature and source files. The earlier full raw-range reconstruction check is preserved as inherited evidence.
- Eight SEEG pairs classify pre-response and rest windows at 70.7% mean balanced accuracy in later trials from eight recordings sci:sha256:c74068384c7c056974c8568aab987cf5a8eee9979a622c22efa1b67a911e77fd
[Legacy relation: reproduces] The current all-band baseline used the same features and chronological splits. A damped Newton solver fit the same C=1 logistic objective. It reproduced the selected pairs, all 1,348 test labels, and all eight balanced accuracies. The largest probability difference from the archived LBFGS fit was 1.68e-6. This is a computational reproduction on reused data.
- 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 used the eight-pair decoder design and its source eligibility rules. Its beta-only and gamma-only results were known before this run. We add a combined low-band comparison with separate pair selection and fixed early training budgets.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
The analysis used all eight patients from OpenNeuro ds004473 version 1.0.2; the archived metadata and raw retrieval records preserve source versions and hashes.
- doi:10.1088/1741-2552%2Facae0a
- doi:10.21105/joss.01896
- doi:10.1038/s41597-019-0105-7
Cited by
7 publicationsPublic catalog entries on Sciscape.
- usesAfter calibration and model fitting, two-bit power features gave 71.2% mean balanced accuracy in eight later-trial SEEG tests
This study supplied the complete feature arrays, trial records, source byte ranges, and decoder helper code. We held its selected-pair rule and test split fixed while we tested feature bit depth.
- usesRecent calibration did not show a clear accuracy gain in eight SEEG recordings with 32 matched calibration trials
We used all preserved feature files, source metadata, raw audit ranges, and baseline model records. We adapted the decoder utilities and source audit code. We used the source extraction and retrieval code to provide a complete rebuild recipe.
- usesAdding 70–110 Hz power gave an uncertain 0.9-point mean accuracy gain after task condition changes in eight records with eight fixed SEEG pairs
Its band test on the same people motivated this fixed-pair transfer comparison. The new test uses less training data and a different test schedule, so it does not directly challenge the earlier estimate.
- usesJoint calibration reduced rest windows with any call across intact and known simulated single contact loss cases in eight offline SEEG tests
This publication supplied the complete SEEG feature arrays, source records, retrieval code, and fixed logistic solver. They enabled the present physical contact loss calibration test.
- usesA scale from other patients gave mean Brier error 0.187 in eight offline SEEG tests
This work produced the same eight-patient SEEG feature data used by the immediate parent. We preserved the inputs and checked selected features against fresh public source ranges.
- usesAn added voltage screen after a key press reduced input coverage in eight SEEG records at the training rest 95th percentile
We used its event windows, three-band power method, adjacent pair rules, and stated 5000 microvolt screen. We tested a new comparison of test selection rules.
- usesEarly confidence scaling decreased mean Brier error from 0.193 to 0.185 in eight offline SEEG patient tests
We use its complete eight-patient SEEG power features, trial exclusions, source records, raw retrieval code, and decoder design. We add a separate early confidence fit and a later probability test.