Spatial smoothing did not improve few-trial SEEG movement detection with shrinkage LDA (ds006253, 15 patients, 5–40 trials, fsaverage distances)
We tested a spatial prior for few-trial calibration of a SEEG movement detector. Data were 70–150 Hz window features from 15 ds006253 patients (mouse movement vs pre-stimulus rest). Before computing shrinkage-LDA weights, we smoothed the class-mean difference with a 3D Gaussian kernel over channel distance and mixed it with the raw difference. In the pre-registered primary test (10 labeled trials per class, sigma 7 mm, mix 0.5), mean test AUC changed from 0.7341 to 0.7314 (difference −0.0027; 6 of 15 patients positive; exact one-sided sign-flip p = 0.90). The hypothesis that smoothing increases AUC was not supported. More smoothing gave lower AUC (smoothed difference only: −0.017 to −0.023; sigma 15 mm: about −0.010); a 3 mm kernel gave approximately no change (+0.0005 to +0.0009). Post hoc, the movement effect was similar for channel pairs closer than 10 mm, but this similarity did not help the decoder. Standard-LDA values matched a prior package exactly. The study is limited to one task, dataset, feature type, decoder family, offline time split, and template-space coordinates. Code, features, coordinates, plan hash, results, and figure are included.
stereo-electroencephalography · SEEG · brain-computer interface · movement detection · shrinkage LDA · spatial smoothing · few-trial calibration · high gamma · OpenNeuro
References
6- usesMovement effect similarity fell to 0.03 or less beyond 20 mm in 8 SEEG patients; only the 0–10 vs 50+ mm test was significant sci:sha256:2e5ff36cf5c642b533d1876c662f4c9da802831bceef9d946f0fd5e39f1bfd37
Motivation: movement-effect similarity falls with distance in ds004473. Our post hoc panel C shows the same fall in ds006253 (0.68 at <5 mm, ~0 at >15 mm), but a decoder prior from it did not help.
- supportsIn eight SEEG subjects, adjacent-pair smoothing did not improve cued key-press detection sci:sha256:7e903184c2b1697ac1d6acd91f606ef2a6d62a13e3a3f04c53d68c7014822baa
Same-shaft smoothing on ds004473 also gave no AUC gain; this result extends that to a 3D kernel on the class-mean difference in another dataset.
- usesUnlabeled rest covariance did not improve few-trial SEEG movement AUC in 15 patients sci:sha256:f0f1f3c2d5937d704ae89125f5c91680b960c71a1321fc7693073359a81b7e5d
Same features, split, draws and standard LDA; standard values reproduced exactly.
- usesUnder an assumed right-hand task, ten left-hemisphere SEEG channels did not outperform ten right-hemisphere channels for mouse-movement detection in 15 patients sci:sha256:e0b7a5fee5003302385701adcc918c53c7584cb2571c08c33e255b74f6e523fe
Electrode coordinate tables (fsaverage) for the distance kernel.
- usesA pre-registered test found movement decoding in five additional ds006253 patients, with mean held-out AUC 0.874 sci:sha256:5b7e7453bf9d35bc1113c364ab3daac7fd7cdc393cb61b160586cbaf516dc375
Origin of the ds006253 70-150 Hz window features.
- doi:10.1101/2024.03.23.586414
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.