Pair selection and fitting each improved mean offline SEEG accuracy with eight pairs and 64 instead of 16 trials in eight recordings
We tested which calibration step gained more from added trials in eight public SEEG recordings. We used OpenNeuro ds004473 version 1.0.2. Each model used eight electrode pairs and three power bands. We crossed 16 or 64 trials for pair selection with 16 or 64 trials for model fitting. The two mixed models each used 64 labeled trials in total. Mean later balanced accuracy was 60.79%, 63.92%, 63.94%, and 67.80%. The settings were 16/16, 16/64, 64/16, and 64/64, in that order. The average contributions from pair selection and model fitting were 3.51 and 3.50 percentage points. The primary difference between the mixed models was -0.015 points. Its descriptive 95% patient interval was -0.94 to +0.84 points. The exact patient sign-flip test gave p = 0.96875. The evidence did not show a larger gain for either step. The result applies to fixed offline response and rest windows from the same recording. The later test was reused from prior work. The package preserves code, features, source records, fitted models, checks, and figures.
stereoelectroencephalography · brain-computer interface · calibration · pair selection · logistic regression · balanced accuracy · offline decoding
References
4- uses64 matched calibration trials increased mean later SEEG balanced accuracy over 16 with a fixed penalty in eight offline recordings sci:sha256:1afe3af4a53f584020860f5028d9cc451d3b18ab565a7a1b0ae0c23074d9dd0c
This work supplied the public trial features, matched calibration samples, fixed penalty, and later test. Its preserved model states let us check both diagonal settings exactly.
- usesEight SEEG pairs classify pre-response and rest windows at 70.7% mean balanced accuracy in later trials from eight recordings sci:sha256:c74068384c7c056974c8568aab987cf5a8eee9979a622c22efa1b67a911e77fd
This work first made the event-based power features and pinned raw retrieval method. We used those features through the parent package and checked raw powers and a fresh source download.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
The analysis used the version 1.0.2 release of the sEEG Forced Two-Choice Task dataset.
- doi:10.1088/1741-2552%2Facae0a
Cited by
1 publicationPublic catalog entries on Sciscape.
- Added fitting trials reduced Brier error more than added selection trials in eight offline SEEG tests with eight pairs and a fixed penalty
[Legacy relation: reproduces] This report already tested the same four main models. We reproduced its samples, pairs, fitted states, and 3,200 main score sets. The same people and trials were used. We added planned Brier and log-loss tests and two fixed penalty checks.