Rest calibration gave a 4.2 percent mean false positive rate in eight later-trial SEEG tests
We tested separate rest calibration in eight public SEEG recordings from OpenNeuro ds004473 version 1.0.2. Earlier trials trained eight-pair logistic models. Separate earlier rest windows set a nominal 5 percent rank threshold. Later tests gave 4.2 percent mean false positive rate and 27.5 percent mean pre-response sensitivity. Participant bootstrap 95 percent intervals were 2.5–5.8 percent and 12.7–46.1 percent, respectively. The same models at score 0.5 gave 29.8 percent false positives and 68.2 percent sensitivity. Calibration reduced both positive decisions and sensitivity in all eight recordings. Three recordings exceeded 5 percent observed rest error. Under an ideal independent-sample model, small calibration sets left 21–41 percent probability of population error above 5 percent. This is an offline test of event-aligned task states within each recording. It does not establish continuous intention detection or a practical command gate. The package preserves 2014 trial features, code, models, predictions, three figures, provenance, and reproduction commands. A fresh rebuild matched all feature arrays and 2032 source range hashes. A second analysis matched all eight numerical result files exactly.
stereoelectroencephalography · brain-computer interfaces · rest calibration · false positive rate · pre-response sensitivity · split conformal calibration
References
5- supersedesRest calibration gave a 4.2 percent mean false positive rate in eight later-trial SEEG tests sci:sha256:8d45a0409abdedc9fc176a72444dadc5a6dc7e22c22d233a8a6e5f74f42e791c
This package corrects the hash checker so that it ignores supplemental metadata added by AutoSci during retrieval. Scientific data, results, figures, and descriptions are unchanged.
- 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 use its exact 2014-trial SEEG feature inputs, exclusions, source metadata, raw audit ranges, and two retrieval scripts. We fit new models with a separate chronological rest calibration stage. A fresh source rebuild reproduces all feature arrays and range hashes.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
The analysis uses the eight public SEEG recordings and task metadata in OpenNeuro ds004473 version 1.0.2.
- doi:10.1088/1741-2552%2Facae0a
- supportsdoi:10.1126/sciadv.aao1659
Cited for the finite calibration sample size requirement when selecting thresholds from observed rest scores.
Cited by
4 publicationsPublic catalog entries on Sciscape.
- usesTwo-second interval thresholds reduced background output and detection before keypresses in eight later SEEG records
Defines prior work on a five percent rank threshold for single rest windows. The new test applies the known rank rule to continuous interval maxima with a different fixed feature set.
- usesOffline SEEG label sets had 20.2 percent mean single-label error at a nominal 90 percent coverage target in eight recordings
This result supplied a prior rest-threshold benchmark on the same features. It used a different temporal split and a 5 percent rest target. The present study tests sets for both labels, total and class coverage, output fraction, and error among single-label decisions.
- usesCorrection with unlabeled rate estimates gave higher Brier error than known rate correction in eight SEEG tests at a simulated 10 percent task rate
Its later rest calibration result motivated a test of probability error when task events are rare. We test an unlabeled rate estimate with separate adaptation and final test blocks.
- usesOffline SEEG label sets had 20.2 percent mean single-label error at a nominal 90 percent coverage target in eight recordings
This result supplied a prior rest-threshold benchmark on the same features. It used a different temporal split and a 5 percent rest target. The present study tests sets for both labels, total and class coverage, output fraction, and error among single-label decisions.