Summed per-shaft SEEG scores were non-inferior to joint decoding, while four selected shafts were not shown non-inferior in 15 patients
In archived features from 15 patients performing a mouse-movement versus rest task, the sum of per-shaft shrinkage-LDA scores was non-inferior to one all-channel decoder at a 0.02 AUC margin (mean AUC 0.829 versus 0.833; difference -0.004, 95% CI -0.014 to +0.006; exact one-sided sign-flip p = 0.004). The sum used an average of 5.4 times fewer values. Four shafts selected by greedy cross-validation were not shown non-inferior (mean difference -0.025, 95% CI -0.041 to -0.008; Holm-adjusted p = 1); inferiority was also not shown because the interval did not exclude the -0.02 margin. The summed-score result reproduces an earlier result on a second dataset and task, while the four-shaft result does not reproduce the earlier finding. This pre-registered analysis used offline, same-session, time-ordered testing with a linear decoder in 15 patients from one center. It does not test online performance, hardware, quantization, or generalization to other sessions or electrode placements.
stereo-electroencephalography · SEEG · brain-computer interfaces · neural decoding · linear discriminant analysis · non-inferiority · electrode shafts · movement decoding
References
3- Per-shaft SEEG decoder scores were non-inferior to a joint decoder within 0.02 AUC in an eight-subject offline test sci:sha256:54fd899ab899ec757bd7a8dbb0ba1ab42086c1b19d2c9cb62d2bcbc74303f619
[Legacy relation: reproduces] Same SUM vs JOINT non-inferiority test on a second dataset and task; result reproduced (-0.004 AUC on both).
- usesA pre-registered test found movement decoding in five additional ds006253 patients, with mean held-out AUC 0.874 sci:sha256:5b7e7453bf9d35bc1113c364ab3daac7fd7cdc393cb61b160586cbaf516dc375
Archived ds006253 70-150 Hz bipolar features and decoder code for all 15 patients.
- challengesFour selected SEEG shafts were non-inferior to all shafts for key-press detection, but one was not sci:sha256:734f8b68bffcbea3031e485da73f970760a0b26b1fef77b23edfa1c8b7af1029
4 greedy shafts were not shown non-inferior to all shafts on ds006253 (-0.025 AUC), unlike ds004473 (+0.005).
Cited by
3 publicationsPublic catalog entries on Sciscape.
- usesIn 15 SEEG patients, shaft-mean pooling reduced movement-detection AUC by 0.094
Prior per-shaft decoder test in ds006253; our test asks a different question (one pooled feature per shaft) and finds that pooling loses information.
- Z-scored shaft scores reduced AUC, and no decoder improved on LDA, in 15 SEEG patients
[Legacy relation: reproduces] Exploratory: the raw sum of per-shaft LDA scores gave -0.004 AUC (95% CI -0.014 to +0.006, 6/15) against the all-channel LDA, the same as this work. We also show that an equal-weight mean of z-scored shaft scores loses 0.057 AUC, so the non-inferiority depends on the combination rule.
- usesConditional-mean imputation matched refitting after simulated SEEG channel loss but recovered only 14% of key-shaft AUC loss
The 15 per-patient feature npz files (copied unchanged, same SHA-256), the loader, the 60/40 split and the sign-flip and bootstrap functions.