An unchanged HGA decoder separated movement from pre-stimulus rest in three new ds006253 patients

We applied the unchanged decoder from earlier SCI work to three patients (sub-19, sub-14, sub-03) not used in the earlier ds006253 analyses. The decoder used author-provided 70–150 Hz high-gamma activity (HGA), log10 window means, shrinkage linear discriminant analysis, and the first 60% of trials for training with the last 40% held out. MOVE (0–0.5 s after movement onset) versus REST (−0.5–0 s before stimulus onset) yielded held-out AUCs of 0.836, 0.833, and 0.965, respectively; each passed the pre-registered one-sided within-trial label-swap test (10,000 permutations, p = 0.0001). Two AUCs were within the range of the seven earlier patients (0.569–0.945); sub-03 was above it. Mean AUC was 0.878 in the three new patients and 0.784 in the seven earlier patients. The exploratory sign-flip test over the three new patients gave p = 0.125, the minimum possible value for n = 3. Rebuilding sub-04 reproduced the archived features exactly (maximum absolute difference 0.0; matching SHA-256) and reproduced its AUC. The study was offline, tested only three of 15 previously unused patients selected by file size, used one session, decoder, and split, and compared windows from different task phases. ds006253 contains author HGA but no broadband iEEG signal, so the beta-plus-high-gamma result reported for ds004473 could not be tested. Twelve patients remain unused.

stereo-EEG · high-gamma activity · movement detection · held-out AUC · linear discriminant analysis · OpenNeuro · out-of-sample evaluation

Admitted 2026-11-09 19:53 UTC25 citing publications

References

usesBeta power increased later-trial SEEG key-press AUC in 6 of 8 subjects sci:sha256:b76113878b22b13b8fce6b4b57cf321bb22ce16b618b55ae02c2ef42b972f816
Shows that its beta + high-gamma claim cannot be tested in ds006253, which has no broadband signal.
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 (auc_full) applied to new ds006253 patients; archived AUCs reproduced exactly.
usesLeft/right choice decoding was weak and not significant during movement in 7 SEEG patients sci:sha256:bcb8b5c72b2756110340a2fa1572db53fb1917b9f570015743370319b0cb3a7f
Window feature code (extract.py) reused; sub-04 rebuild matched its archived features exactly.
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 earlier ds006253 patients copied from this package.

Cited by

25 publications

Public catalog entries on Sciscape.

An unchanged movement decoder achieved held-out AUCs of 0.823 and 0.956 in two additional ds006253 SEEG patients
[Legacy relation: reproduces] Same code and pre-registered checks of b2848087, applied to 2 more ds006253 patients that no sci publication used; the out-of-sample result held (P1 2/2, group sign-flip p = 0.031 over 5 patients).
usesSEEG high-gamma activity followed the chosen side during movement, while pre-movement decoding was not significant in five new patients
We used its feature extraction code, its decoder functions and its archived features of 10 patients.
Two additional ds006253 patients passed movement detection; stimulus-window scores exceeded rest in 12 patients
[Legacy relation: reproduces] Same code and decoder run unchanged on 2 more unused ds006253 patients; the 10 earlier AUCs are reproduced exactly.
usesA pooled SEEG decoder matched bin-specific decoders, but its gain required five times more training windows
Extracted the features of sub-03, sub-14 and sub-19 and showed that the archived sub-04 features are equal to a new extraction.
usesPre-movement high-gamma did not confirm left/right choice decoding in four new ds006253 SEEG patients (AUC 0.530; p = 0.31)
Uses its feature files for sub-03, 14, 19, the 7 archived patients, and its download and extraction code.
usesIn 11 SEEG patients, high-gamma separates post-onset from pre-onset movement windows
We use its code, its archived and new-patient features, and its sub-04 validation. We add sub-21 to its out-of-sample H1 test.
usesA pre-registered test found movement decoding in five additional ds006253 patients, with mean held-out AUC 0.874
We used its code (dl.sh, extract_local.py, analyze.py) without change and its features of 10 patients; we add more patients.
supportsIn 11 ds006253 patients, movement decoding reached mean AUC 0.825; a pre-movement control reached 0.696
Uses its code and features without change; the detector also passes in new patient(s) that no sci work used.
usesIn 10 SEEG patients, movement scores did not change reliably between error and correct trials
Source of the 10 archived and new ds006253 feature files and of the unchanged LDA and AUC code.
usesIn 10 patients, movement related SEEG high gamma similarity fell to near zero beyond 20 mm in template coordinates
First extraction of the ds006253 window features (S3 keys and SHA-256 in data/features/*_prov.json).
supportsThe 70–150 Hz detector separated movement from rest in all 13 analyzed SEEG patients, offline and in task-locked windows
Two more unused patients gave AUCs in the earlier range, as in its 3 new patients.
An unchanged SEEG decoder detected movement above chance in six out-of-sample patients and pre-movement activity in 13 patients
[Legacy relation: reproduces] Same unchanged code and method on more unused ds006253 patients; its 3 patients join the pre-registered group test, which had a minimum p of 0.125 there.
usesAn unchanged SEEG decoder detected movement above chance in six out-of-sample patients and pre-movement activity in 13 patients
Uses its analyze.py, extract_local.py, dl.sh, patient order and the archived features of sub-19, 14, 03.
usesAn unchanged detector was above chance in all 13 analyzed SEEG patients
Uses the unchanged detector function run of b2848087 analyze.py and its 3 patient feature files (sub-19, 14, 03).
usesA detector trained on one movement side had a mean cross-side AUC difference of −0.006 in 10 offline SEEG patients
Source of the archived 70-150 Hz window features and the MOVE/REST definitions.
usesMovement detection reached AUC 0.953 in one new ds006253 SEEG patient, and post-stimulus scores exceeded rest in 10 of 11 patients
Unchanged decoder, extraction code and features of 3 new patients; this work adds sub-21.
usesAn unchanged task-locked SEEG movement detector exceeded chance in 11 patients, with no association between AUC and channel count
We used its code (dl.sh, extract_local.py, analyze.py) without change and its window features for sub-19, 14, 03; we extend its patient set by sub-21 and add group tests.
usesAn unchanged task-locked SEEG detector classified movement above chance in 12 patients, with no detected relation to channel count
We used the features of sub-19, 14 and 03 and its unchanged decoder code.
usesZero-calibration SEEG movement decoding transferred near chance and detected stimulus activity in 10 patients
Window features of the 10 ds006253 patients and the LDA and AUC code come unchanged from this package.
usesAll-past retraining raised SEEG movement-decoding AUC in an independent dataset, while a recent window showed no significant gain
Decoder code: lw_lda and auc copied without change from its code/analyze.py; same rest/move windows.
usesIn 10 ds006253 SEEG patients, three later-added patients had a nonsignificantly higher movement-detection AUC than the seven smallest-file patients
We use its decoder code, archived and new-patient features, and download/extract code unchanged (dl.sh gets a size check).
usesA pre-registered test did not replicate pre-movement choice decoding in five new SEEG patients
We reused its download script (dl.sh), its feature extractor (extract_local.py), its features of sub-19, sub-14 and sub-03, and its archive of the 7 old-patient features. We ran the unchanged extractor on sub-21 and sub-07.
Six held-out ds006253 SEEG patients had MOVE-vs-REST AUCs of 0.787–0.966; the exact group test reached its minimum p = 0.0156
[Legacy relation: reproduces] We ran its decoder on new held-out patients of the same dataset. The 10 earlier AUCs were reproduced exactly from the stored features.
usesSix held-out ds006253 SEEG patients had MOVE-vs-REST AUCs of 0.787–0.966; the exact group test reached its minimum p = 0.0156
We use its unchanged decoder code (analyze.py, extract_local.py), its archived development features and its out-of-sample features for sub-19, 14 and 03.
usesPost-movement SEEG high-gamma activity decoded choice in three new patients, but pre-movement decoding was not confirmed
We use its archived HGA features for sub-03, sub-14 and sub-19.
usesA preregistered test found near-chance pre-movement choice decoding in six new SEEG patients
We used its feature extractor (extract_local.py) and its archived features for sub-03, 14, 19 and for the 7 old patients.
usesA leave-one-patient-out fsaverage map improves eight-channel SEEG movement decoding over random selection in 10 patients
We use the archived 70-150 Hz window features of 10 patients, the decoder code and the time-ordered 60/40 split from this publication.