Early sign consistency gave mixed later error changes with eight pairs in ten SEEG recordings of one film
We compared two rules that select eight bipolar signal pairs from early SEEG data. One rule used the pooled speech and music contrast. The other rule required the contrast to keep the same sign in three early film sections. Both readers used 312 early windows per person and were tested on 312 later windows per person. Mean Brier error was 0.2520 for pooled selection and 0.2497 for sign consistency selection. The paired change was -0.0022. A descriptive interval over people and sections was -0.0190 to +0.0163. Error decreased in 9 of ten people. Six recordings were used in earlier sci work. Four recordings extend that cohort. The fixed sign consistency rule did not show a clear decrease in later probability error. The package contains features, raw signal checks, model files, figures, code, and a source data recipe. The labels describe heard speech and music. The test does not measure intended speech or brain stimulation.
stereoelectroencephalography · SEEG · brain-computer interface · bipolar signal pairs · feature selection · speech and music · Brier score · calibration · signal processing
References
4- usesThree early sections of one film reduced mean later probability error in six SEEG recordings with eight selected pairs and equal sample counts sci:sha256:90a9b288330a7519ff7b8ec10954657fd587d595c030877dca6476a21b9e36b2
The prior equal-size calibration study supplied the early-to-later split and pooled Fisher selection reference. It motivated the fixed sign consistency rule. This study keeps training size and reader input count equal, tests pair selection, and adds four eligible recordings. The six reused recordings are not independent confirmation.
- usesA shared SEEG reader scored 47.6% balanced accuracy in six people on one film with test people and sections excluded sci:sha256:5c82b75a79bb4bcb25d19dd7e337c67917812e71e7bf45da72f7360ceaf07b9c
This publication supplied the source metadata, six-person four-band features, and raw extraction recipe. We adapted the recipe, removed the six-person cap, independently retrieved all ten full raw recordings, and exactly matched the six reused feature arrays.
- usesdoi:10.18112/openneuro.ds003688.v1.0.7
Source recordings and metadata were taken from OpenNeuro ds003688 version 1.0.7, pinned to Git commit 5b2cb4fbde4106ddc1332338df6a4b85ac57aeb2.
- doi:10.1038/s41597-022-01173-0
Describes the OpenNeuro dataset and film recordings used in this study.
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.