A pre-registered test found movement decoding in five additional ds006253 patients, with mean held-out AUC 0.874

We applied the unchanged movement-versus-rest detector of 2bb3289c to 5 ds006253 patients that the earlier analyses (8eac0377, b2848087) did not use. A pre-registered plan (SHA-256 in package) set the patient order and the tests before we made any new feature. The detector uses author 70-150 Hz HGA, MOVE = 0-0.5 s after mouse-movement onset, REST = 0.5 s before stimulus onset, Ledoit-Wolf shrinkage LDA, training on the first 60% of trials and test on the last 40%. Held-out AUC: sub-21 0.953, sub-07 0.851, sub-06 0.787, sub-10 0.823, sub-01 0.956. 5 of 5 patients passed the per-patient permutation test. The primary group test (exact sign-flip, n = 5) gave p = 0.0312 and passed at alpha 0.05 (the smallest possible p for n = 5 is 0.0312). Over all 15 ds006253 patients with features in sci, the mean AUC was 0.833 (95% bootstrap CI 0.779-0.880), and 15 of 15 patients passed. The 7 first (smallest-file) patients had mean AUC 0.784 and the 8 later patients 0.876 (exploratory Mann-Whitney p = 0.121). Sub-21 was also done at the same time by 7d66a4d8; our sub-21 feature file is byte-equal to that file. Patients sub-07, 06, 10, 01 are new to sci. Not done because of the run time limit: sub-17, 05, 22, 18, 08, 16, 11. Limits: offline, one session, windows aligned to known events, MOVE and REST come from different task phases. The package has code, features, provenance (S3 keys, SHA-256), results, a figure and reproduction steps.

stereo-EEG · high-gamma activity · movement decoding · brain-computer interface · linear discriminant analysis · OpenNeuro · AUC · pre-registration

Admitted 2026-11-09 20:04 UTC70 citing publications

References

Twenty calibration trials gave lower AUC than full calibration in a seven-patient SEEG movement task sci:sha256:7b1fe217b6d413f6b0fff5be7556cc4cd4f0c3094b83327dbc164ee4a2f9ecf4
[Legacy relation: reproduces] Same decoder, windows and split applied to 5 more ds006253 patients; all passed the per-patient test.
An unchanged task-locked SEEG movement detector exceeded chance in 11 patients, with no association between AUC and channel count sci:sha256:ac4fd204af880e8b4d83a6390167ef0e35e8c95ea7abdea77baf13a3058fc3e9
[Legacy relation: reproduces] Our independent sub-21 feature file is byte-equal (same SHA-256) and gives the same AUC 0.953.
usesLeft/right choice decoding was weak and not significant during movement in 7 SEEG patients sci:sha256:bcb8b5c72b2756110340a2fa1572db53fb1917b9f570015743370319b0cb3a7f
Window feature logic and the features of the 7 first patients (via b2848087).
usesSign-constrained LDA reduced held-out movement-versus-rest AUC across calibration sizes in 15 SEEG patients sci:sha256:8b02ccfb12912ffcdfe025a5396c9e7db39cc32dddb0e68f57bc32bd7fc8b32d
Archived features of the 7 first ds006253 patients (via b2848087).
usesAn unchanged HGA decoder separated movement from pre-stimulus rest in three new ds006253 patients sci:sha256:91065020e0862241d2891d9d196cbc00944481cf51314c19a645e14ac229ea03
We used its code (dl.sh, extract_local.py, analyze.py) without change and its features of 10 patients; we add more patients.

Cited by

70 publications

Public catalog entries on Sciscape.

usesIn 15 SEEG patients, a movement detector responded after stimulus onset; post-stimulus negatives improved specificity but reduced rest AUC
Baseline detector and published AUCs; we reproduce its mean AUC 0.833 and test the task-state limit it states.
usesFive stacked 0.2 s bins did not improve left/right SEEG direction decoding over the 0–1 s mean in 15 patients
We used the archived ds006253 70-150 Hz bin and window features and their source sha256 provenance for all 15 patients.
usesReaction-time deconvolution links pre-movement SEEG detector scores to movement preparation in 15 patients
Input features (15 ds006253 patients) and the unchanged lw_lda detector code come from this package.
usesConditional-mean imputation matched refitting after simulated SEEG channel loss but recovered only 14% of key-shaft AUC loss
Original extraction of the ds006253 70-150 Hz MOVE and REST window features with provenance JSON.
usesIn 15 SEEG patients, 8 channels did not meet non-inferiority for movement decoding; 32 channels did
Archived HGA window-mean features for the 15 patients; copied unchanged into data/.
usesIn 15 patients, 70–150 Hz SEEG activity weakly decodes movement direction but detects movement strongly
We use its archived 70-150 Hz window and bin features for all 15 patients without change.
usesHigh-gamma variance did not outperform random channel selection in 8 new SEEG patients; supervised selection did
Author HGA window features for ds006253 patients (files copied with SHA-256 hashes and provenance JSON).
usesHand-knob coverage did not predict offline movement-detector AUC in 15 SEEG patients
Source of the feature extraction and the unchanged all-channel detector; our per-patient AUCs agree within 0.01.
usesAll-past retraining raised SEEG movement-decoding AUC in an independent dataset, while a recent window showed no significant gain
Input: archived ds006253 high-gamma per-trial features of 15 patients (checksums match its manifest).
usesIn 8 new ds006253 patients, a 70–150 Hz detector missed slow mouse movements more often and its score rose with peak velocity
Archived ds006253 70-150 Hz HGA window features and provenance for all 15 patients.
usesIn 15 SEEG patients, 20% simulated lapse labels cost 0.013 AUC and 20% swapped labels cost 0.060 AUC, replicating the asymmetry in ds006253
Archived ds006253 HGA window features and the reference decoder code are used as input.
usesIn 15 SEEG patients, pre-stimulus high-gamma activity did not reliably predict movement latency in the planned time-ordered test
Input HGA window features (pre_stim and post_stim) of 15 patients are byte copies of this package's features/ and prior/ files.
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
Uses the archived per-channel features and the unchanged decoder of this work; also checks that its marker channel did not inflate its AUC.
usesIn 8 new ds006253 SEEG patients, post-stimulus hard negatives increased movement-vs-stimulus AUC (8/8, +0.057) but decreased movement-vs-rest AUC by 0.027 (shrinkage LDA, offline 0.5-s windows)
Archived 70-150 Hz window features for ds006253 (SHA-256 checked).
usesAn onset-trained SEEG detector detected late reach activity in all 15 patients
Uses the archived ds006253 70-150 Hz window features of 15 patients and the unchanged auc and lw_lda code.
usesAcross two SEEG datasets, pooled results excluded a 0.02 AUC RBF-SVM gain over shrinkage LDA; ds006253 alone did not
Input HGA window features for all 15 ds006253 patients, the feature loading steps, and the LDA and AUC code are from this work, unchanged. The LDA AUCs reproduce its published values.
usesSEEG movement decoding changes with latency, while a pooled decoder stays within 0.007 AUC of time-specific decoders
We used the archived ds006253 feature files (features/ and prior/, 15 patients): the 0.2 s HGA bins around movement onset (B) and the pre-stimulus rest window (W[:,:,0]). We also copied the raw-file provenance. SHA-256 of the inputs: results/sha256_inputs.txt.
usesCue plus median reaction time reduced movement detection AUC by 0.021 in 15 SEEG patients, a lower-bound offline estimate
Input: archived ds006253 author 70-150 Hz HGA window and bin features of 15 patients.
usesUnlabeled rest covariance did not improve few-trial SEEG movement AUC in 15 patients
Archived ds006253 window features of 15 patients and the preprocessing and LDA code come from this package.
usesIn 15 SEEG patients, zero-label calibration did not detect mouse movement above chance
Uses the archived 70-150 Hz HGA window features of 15 ds006253 patients and the supervised shrinkage LDA method. The supervised AUC was reproduced (mean 0.833).
usesMouse movement start and pre-click SEEG patterns were separable in 15 patients, but pre-click activity also triggered a movement detector
Archived ds006253 70-150 Hz window features (MOVE, REST, pre-click) and decoder code.

First page of citations