Simulated noisy SEEG channels slightly reduced recall-decoder AUC; repair did not improve the primary AUC test
In an offline simulation using archived FR1 band-power features from 18 SEEG patients (OpenNeuro ds004789 v3.1.0), we added a +5 calibration-SD shift and Gaussian noise (SD 2) to 1, 4, or 16 bipolar channels after decoder calibration. Mean AUC fell by 0.0010 for 4 channels and 0.0046 for 16 channels. The pre-registered primary test found no significant AUC increase after label-free repair at 4 channels (mean difference +0.0006; one-sided exact Wilcoxon p = 0.34; Holm p = 0.67). At +5 SD the first-12-word detector found all noisy channels and produced no false flags in the simulated runs. Exploratory fixed-threshold balanced accuracy fell by 0.012 and 0.039 at 4 and 16 channels; repair recovered it to within 0.001–0.003 of clean. These results are limited to simulated faults, one dataset and session, 18 patients, and one strongly regularized L2 decoder.
SEEG · brain-computer interface · word recall · noisy channels · fault detection · label-free repair · AUC · balanced accuracy · simulated faults
References
4- supportsIn 18 offline SEEG patients, refitting partly recovered AUC after 50% random channel loss; shaft loss caused greater decline sci:sha256:e8c2892080a762f7780aaae9df84f7018a351b41afb3397a55132fa5b63805fd
It found that channel loss after calibration has a small AUC cost for this decoder. We found that noisy (not removed) channels after calibration also have a small AUC cost (0.001-0.005).
- supportsIn 18 SEEG patients, clipping band-power features at 3 SD slightly increased recall AUC, while rejecting 5% outlier training words did not sci:sha256:13958e32609a0456549eb4db9a86bcb83fd62e2d75accfb613c12b51177b570a
It found that clipping outlier features gives little AUC gain without faults. We found that a label-free repair of simulated noisy channels also gives little AUC gain, but it recovers fixed-threshold balanced accuracy.
- usesCross-patient SEEG recall decoding was above chance but weaker than a three-list within-patient decoder sci:sha256:d52ba56c23a69304d837e4ef5a9f3f20bcee59ebf2cd093faea38a0334a9fa4a
We used its archived ds004789 FR1 feature files, its inclusion rule, its W_chan decoder settings (C = 7.2e-4) and its solver, AUC and Wilcoxon code without change.
- usesdoi:10.18112/openneuro.ds004789.v3.1.0
The archived FR1 session 0 band-power features used in this analysis are identified as OpenNeuro ds004789 v3.1.0.
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.