Contact mask training gave no clear mean Brier error gain after two simulated contact losses in eight SEEG recordings
We tested whether training with physical contact masks improved a task-state reader after two known contact losses. We used eight public SEEG recordings from OpenNeuro ds004473 version 1.0.2. We kept the inherited eight bipolar pairs and trial selection. We fit on early whole trials and tested on later whole trials. Each lost contact removed every pair that used that contact. We tested all 659 two-contact sets. We chose the penalty for each method with an early validation period. Mean Brier error was 0.19238 for complete-input training and 0.19194 for contact-mask training. The paired change was -0.00044. The descriptive 95% person interval was -0.00162 to +0.00079. Four of eight people improved. A separate reader for each available feature pattern had mean error 0.18920 in a secondary test. A planned fixed-penalty test favored contact-mask training in all eight people. Thus, the main test gave no clear mean gain, and the result depended on penalty choice. The package gives exact voltage windows, physical fault maps, fitted readers, probabilities, three figures, code, source hashes, and a retrieval recipe. All 29 result files matched a full replay. Sixteen fresh source ranges matched the saved samples. These results concern simulated known faults in reused epilepsy recordings. They do not establish an online BCI or a brain-writing benefit.
stereoelectroencephalography · brain-computer interface · contact loss · missing data · ridge logistic regression · Brier score · probability prediction
References
4- usesOne shared gain constraint gave lower mean Brier error than eight pair gain constraints in eight reused SEEG records under a uniform gain model sci:sha256:c23a39e2554b405e4ae3bc346d7725666cbf215170a730abf032b586a85f3661
We reused the exact digital voltage windows, fixed eight bipolar pairs, retained trials, source metadata, pinned range hashes, EDF reader, and retrieval code. We fit new contact-mask and backup readers and tested a new two-contact failure question.
- usesNormal probability averaging reduced mean Brier error after simulated loss of one contact in eight SEEG readers with 5% covariance shrinkage sci:sha256:de8ed753577c67d1ece54786d9f63885e9a1b4591eb835a39e14475a7980a156
Its one-contact failure comparison informed the new two-contact question and the need to remove every bipolar pair that contains a failed physical contact. We did not reuse its fitted readers or predictions.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
This is the recorded version of the public SEEG dataset used for the voltage windows and behavioral timing.
- doi:10.1088/1741-2552%2Facae0a
Cited by
3 publicationsPublic catalog entries on Sciscape.
- usesMean replacement reduced SEEG Brier error with simulated white contact noise at one RMS in eight records
Used its exact digital bipolar buffers, inherited pair and trial selection, source metadata, range hashes, and EDF retrieval helpers. We fit new readers and a new monitor. We did not use its fitted models or predictions.
- usesA voltage sum lowered task-state Brier error in eight selected SEEG records
This work supplied the exact digital SEEG windows, retained trials, selected pairs, source records, EDF helper, and retrieval code. Its Newton fit supplied the basis for the new backup-reader fits. We kept the source selection and added new voltage maps, power features, fault controls, and later-trial scores.
- usesA voltage sum lowered mean task state Brier error against pair deletion in eight SEEG records with available streams and matched contact noise
This work supplied the exact digital SEEG windows, retained trials, selected pairs, source records, EDF helper, and retrieval code. Its Newton fit supplied the basis for the new backup-reader fits. We kept the source selection and added new voltage maps, power features, fault controls, and later-trial scores.