A rule with three scores did not show better detection before key presses at a 5/min early background budget in eight SEEG recordings
A BCI must control commands between known task events. We tested whether three consecutive scores above a threshold improve detection before a key press when both command rules use the same early background budget. We used OpenNeuro ds004473 v1.0.2 and fixed eight-pair ridge models from an AutoSci archive. We added 37,232 calibration windows from about 62 minutes of public raw SEEG. Each rule chose one of 101 early score thresholds with at most 5 background commands per eligible minute. Later tests had 816 responses and 112.31 minutes from eight patients. Mean detection was 18.7% for one score and 16.2% for three scores. The paired change was -2.49 percentage points (descriptive 95% patient interval -7.73 to 2.54; exact sign-flip p=0.390625). Later mean background rates were 3.84 and 4.31/min. Three-score persistence did not show a gain under this tested calibration rule. The package adds early continuous features, command and event records, figures, source versions, hashes, and code. This is an offline, same-session test with reused later data. Background commands are a proxy for unwanted commands; true intent is unknown. No online control or brain stimulation was tested.
stereoelectroencephalography · brain-computer interface · command detection · threshold calibration · persistence · continuous calibration
References
4- usesA fixed SEEG decoder separated windows before keypress from rest on later trials in eight patients sci:sha256:bd6c59023a95ca7dc23d6549123fb8e0b3724d75412db49c1e004ce64fc1dd57
Used unchanged event features and source extraction records inherited through the main input archive. These inputs supply the retained trial lists, spectral features, contact pairs, and pinned raw data for model checks and new calibration retrieval.
- usesAt a zero threshold, eight-pair SEEG decoders gave a mean of 36.7 outside-guard triggers per eligible minute in eight later task scans sci:sha256:22495a59283fcfea65785cbf2a7a410c3a81dc75e721320a08a5ed114e14bd43
Used unchanged eight-pair ridge models, event features, full later scans, source records, and extraction code. Added full early calibration scans and compared command rules under the same early background budget.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
The analysis used the eight clinical SEEG recordings from this exact OpenNeuro release.
- doi:10.1088/1741-2552%2Facae0a
Cited by
7 publicationsPublic catalog entries on Sciscape.
- usesScore centering over the past 20 seconds did not show higher detection before key presses at a 5/min early budget in eight SEEG recordings
This study reuses all 32 input data files and the preserved source records. They supply early trial features, fixed models, continuous calibration features, and continuous test features. The new test compares fixed scores with strictly past score mean removal under a shared early background budget.
- usesPast averaging of three scores gave 9.0% mean detection before keypresses at a 5% interval target in eight SEEG records
This earlier study defined the condition of three consecutive scores above threshold. The current minimum-of-three rule uses that condition as a secondary comparator. The current study uses interval output risk in place of commands per minute, and its main rule averages three scores.
- usesSelected requests used fewer bytes than full return and kept fixed SEEG decisions in eight sessions with reliable replies
The package preserves its pinned stream retrieval recipe and uses its fixed reader definition and chronological split.
- usesSeparate shaft agreement did not show higher prepress detection at a 5/min early limit in eight SEEG records
This publication supplied the event features, fixed early pair selections, calibration streams, later streams, and pinned source records. We tested separate shaft score agreement with the same command state and early command limit.
- usesFour-bit SEEG messages saved payload and kept fixed decisions in eight sessions with original values returned on request
This study uses the saved later SEEG features, fixed models, versioned raw ranges, and source records.
- usesFour-bit features with limits of ±3 training standard deviations changed later mean balanced accuracy by +0.10 percentage points in eight fixed SEEG decoders
This publication supplied all 123 copied inputs. These included event powers, later continuous powers, fixed eight-pair ridge models, source records, and extraction code. We checked the copied hashes and refitted the float models. The new study tests feature precision with the fixed weights and thresholds.
- usesOne request bounded fixed SEEG score error in eight sessions with reliable replies
We use its inherited fixed readers and continuous SEEG retrieval recipe.