A distance monitor missed 54.4% of new errors on average under simulated ±6 dB shaft gains in eight SEEG recordings
We tested two fixed signal monitors on public SEEG data from OpenNeuro ds004473 version 1.0.2. Early trials set eight-pair task decoders, signal models, and alarm thresholds. Later data supplied 674 test trials from eight recordings. Each trial had a rest window and a pre-response window. Simulated common voltage gains changed one selected shaft group at a time. At plus or minus 6 dB, distance monitoring detected a mean of 45.6 percent of new decoder errors. It missed 54.4 percent. It left new errors without a flag in 9.01 percent of all windows. Separate power limits left 10.54 percent. The paired mean difference was 1.53 percentage points. The descriptive 95 percent patient bootstrap interval was 0.55 to 2.57 points. The exact sign-flip p value was 0.03125 under independent-patient and symmetric-null assumptions. Distance monitoring rejected 45.8 percent of windows. An extra descriptive check found more missed errors than expected from random rejection at the same flag rate. All 2014 input trials matched a fresh raw source rebuild. The package preserves features, provenance, source code, models, all test scores, and four figures. This exploratory result uses a previously studied cohort. It tests simulated gain faults and sampled task states. It establishes no effect on live BCI control or brain writing.
brain-computer interfaces · stereoelectroencephalography · signal monitoring · decoder errors · gain changes · Mahalanobis distance · power features · fault detection
References
3- usesEight-pair SEEG reserve decoders gave 61.7% mean balanced task accuracy after simulated loss of all primary shafts in eight recordings sci:sha256:3c2cc13a3310a10095ea01ce382c47a1a29f11644ef5afdfa7dae05c09bd9eb9
We copied the eight feature files, source metadata, raw audit ranges, and unchanged extractor from this package. They supplied the input and source retrieval recipe for the new hidden-gain monitor test. We freshly rebuilt all retained trial features.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
The study uses public SEEG recordings, event and channel metadata from OpenNeuro ds004473 version 1.0.2; the saved source manifest identifies this exact release.
- doi:10.1088/1741-2552%2Facae0a
Cited by
3 publicationsPublic catalog entries on Sciscape.
- usesTraining for ±6 dB gain on one shaft raised correct classes with a guarantee in eight SEEG records under a linear gain model
We used this package for the eight-person cohort, exact power features, chronological split, pair rule, source range recipes, extraction code, and standard logistic solver. We checked all parent files and rebuilt all source features. We compare new gain training with its standard fit.
- usesA distance gate rejected a mean of 99.96% of inputs with a simulated 1 mV, 10 ms contact pulse in eight SEEG recordings
This report supplies context for distance monitoring under a different simulated fault in the same cohort. We read it after fixing the present plan. It supplies no data or model values to this analysis. Different gates, faults, and splits prevent a direct fault-type comparison.
- usesA shaft loss detector found more simulated −6 dB losses than a distance detector in eight SEEG records
The earlier distance study found near-complete rejection at large gain changes. This result motivated the directed loss comparison and the saved change of the primary gain from minus 12 dB to minus 6 dB before new scores. We fitted both methods again on a common new split.