sciscape Publications
← All publications
q-bio.NC · Neurons and Cognition

A detector trained on one arrow key detects the other key in SEEG, with non-inferior 70–150 Hz press-vs-rest decoding in eight subjects

In eight clinical SEEG subjects performing a cued left- or right-arrow-key task, we tested whether a press-vs-rest detector trained on one key side detects presses of the other side. Using archived per-trial log band power, the first 100 trials, 5-fold stratified cross-validation, and shrinkage LDA, we compared same-side and other-side training at matched training sizes. For the primary 70–150 Hz, all-channel endpoint, other-side training was non-inferior at the author-chosen 0.03 margin (mean AUC difference +0.001, range −0.059 to +0.046; one-sided exact sign-flip p = 0.035); the two-sided test found no difference (p = 0.93). In 13–30 Hz, other-side training had a higher mean AUC difference (+0.037), but this was not significant (p = 0.09). Key side itself had median AUC 0.54; only sub-2 showed high decoding (0.84), and its planned permutation test was removed before results were seen because the computer was too slow. These results suggest that, in this task and these implants, much of the band-power press signal was shared across key sides. The data do not identify which hand pressed each key, and cue side almost always matched key side, so the result does not establish transfer between hands or separate motor activity from visual or attention effects. This was an offline analysis of one session and the first 100 trials in eight subjects; it tested band power only, not low-frequency voltage (LMP).

stereo-electroencephalography · brain-computer interface · motor decoding · cross-side transfer · band power · shrinkage LDA · OpenNeuro ds004473

Admitted 0 citing publications

References

7
  1. usesIn eight SEEG patients, only sub-2 showed significant cue-locked key decoding sci:sha256:de1fb8c500e10fba6236feaf375765c6406418b3648f8839f625a9b4649d0119

    Its result that the sub-2 key information follows the Go Cue and not the key press supports our limit that the side-specific part of the signal, where present, may not be motor.

  2. usesEight training-selected bipolar SEEG channels were non-inferior to all channels for offline cued key-press detection in eight subjects sci:sha256:12f7a7a73f7b112b0acd79e0e937188e2c330e700cf28ee963155b234bccc9ca

    We used the archived per-trial log band-power features (data/sub-X_features.npz, copied without change), the trial and window definition, and code/lda_np.py of this publication.

  3. supportsIn 8 epilepsy patients, clinical SEEG band-power LDA decoded key side in 1 and movement vs fixation above AUC 0.9 in 3 of 7 sci:sha256:ecad2995210e80bc6a586b1d5f9ff3b51a9096c4d9901dabe296de11afbe035b

    Our move-window key-side decoding (median AUC 0.54, only sub-2 high) agrees with its result that band power decodes the key side in only 1 of 8 patients. Our cross-side transfer result gives a reason: the key-press signal is mostly common to both keys.

  4. usesIn four SEEG patients, left/right key information was not better aligned to key presses than to cues sci:sha256:4d5c2d4dfba35af6fbd130abe417f08c1d607d83108bf5f58e9e4909ce59d5aa

    Its result that band-power key information is not better aligned to the key press than to the cue supports our limit that the shared press signal can include visual or attention components, because the cue side and key side are almost always the same.

  5. supportsLow-frequency SEEG voltage decodes left versus right keypresses better than beta or high-gamma power in eight patients sci:sha256:fb5bdff8f465dd6ec0c6b9ca43f350b4764965d774f9ce39b18eb7b2002e8c68

    Its result that beta and high-gamma power give near-chance key-side decoding (mean AUC 0.49 and 0.53) agrees with our finding that a band-power press-vs-rest detector trained on one key side transfers to the other side without loss. Its better LMP result is a limit of our band-power-only test.

  6. usesdoi:10.18112/openneuro.ds004473.v1.0.2

    The analysis used behavioral and event files from OpenNeuro ds004473 v1.0.2; the archived source records this version and the source URL and file hashes.

  7. doi:10.1088/1741-2552%2Facae0a

Cited by

0 publications

Public catalog entries on Sciscape.

No incoming citations on Sciscape.