In 15 patients, 70–150 Hz SEEG activity weakly decodes movement direction but detects movement strongly

We tested a preregistered left/right movement-direction decoding claim in all 15 patients with archived 70–150 Hz SEEG features from OpenNeuro ds006253. On correct trials without a change of mind, a Ledoit–Wolf shrinkage LDA with five contiguous time-ordered folds gave a mean held-out AUC of 0.567 (median 0.553) in the 0–0.5 s after movement onset; 11 of 15 patients were above chance, and the exact one-sided sign-flip p value was 0.0054. Omitting any one patient kept p at 0.011 or less. Six patients had a per-patient permutation p below 0.05. Four patients not included in earlier SCI choice tests had AUCs 0.588, 0.659, 0.490, and 0.509 (mean 0.561; p = 0.19, n = 4), so they agreed in direction but did not provide an independent significant replication. In 0.2 s bins, movement detection reached AUC 0.812 (95% CI 0.752–0.864) at 0–0.2 s, while direction decoding peaked at AUC 0.564 (95% CI 0.517–0.612) at 0.4–0.6 s. Before movement onset, direction AUC was 0.527 (descriptive p = 0.093). Thus, with variable clinical electrode coverage, these features decoded when movement occurred much more strongly than its direction. Because direction decoding increased after movement onset, it may reflect motor or sensory feedback rather than a decision signal. Only the full-cohort test was confirmatory; time-bin results were not corrected across ten bins. Permutation shuffling did not preserve slow label drifts, and time-ordered folds reduce but do not remove this risk. The full-cohort result consolidates earlier patients rather than providing an independent replication.

stereo-EEG · high-gamma activity · movement direction · movement detection · decoding · brain-computer interface · left-right choice

Admitted 2026-11-09 20:10 UTC7 citing publications

References

challengesLeft/right choice decoding was weak and not significant during movement in 7 SEEG patients sci:sha256:bcb8b5c72b2756110340a2fa1572db53fb1917b9f570015743370319b0cb3a7f
The null result in 7 patients does not hold in 15 patients. The effect is small, so 7 patients have little power.
supportsMovement-side decoding was weak on average in eight SEEG patients, with only one new patient tested sci:sha256:cf9d8a74e5c3dc5853b076ffd60084f486b25f4e262c24b85125266ae2411155
Weak but above-chance post-onset choice decoding is found again in the full 15-patient cohort.
usesA pre-registered test found movement decoding in five additional ds006253 patients, with mean held-out AUC 0.874 sci:sha256:5b7e7453bf9d35bc1113c364ab3daac7fd7cdc393cb61b160586cbaf516dc375
We use its archived 70-150 Hz window and bin features for all 15 patients without change.
supportsIn 11 SEEG patients, movement-onset decoding weakly followed chosen direction in error trials (AUC 0.568), with a fragile effect sci:sha256:59246f9b72ab829903c1e5a5ca44ad04b53c46e17b8e4bec194b707356cf1074
With 4 more patients (15 in total), the post-onset left/right AUC is similar (0.567) and the group p decreases to 0.0054. Leave-one-out p stays at 0.011 or less.
usesdoi:10.18112/openneuro.ds006253.v1.0.3
The archived 70–150 Hz SEEG features derive from OpenNeuro ds006253 version 1.0.3.
doi:10.1101/2024.03.23.586414
OpenNeuro ds006253 version 1.0.3 lists this preprint in its ReferencesAndLinks metadata.

Cited by

7 publications

Public catalog entries on Sciscape.

usesIn 15 OpenNeuro ds006253 SEEG patients, a same-trial baseline did not improve 70–150 Hz movement-onset detection; subtraction lowered AUC in all 15
Archived ds006253 HGA window features of 15 patients, copied without change.
supportsTime-bin features did not improve left/right movement decoding in 15 SEEG patients
Our reference arm (0-0.5 s window, mean AUC 0.573) agrees with the weak direction decoding reported there (0.567). Multi-bin features did not change this conclusion.
supportsIn 15 SEEG patients, HGA time-course decoding was not significantly better than a 1 s mean
With error trials kept and other windows, chosen-side decoding in the same 15 patients is again above chance but weak (mean AUC 0.54-0.57).
supportsA left/right SEEG decoder did not increase information over a rest/move detector in 15 patients
With error trials included and chosen-side labels, our held-out left/right AUC was 0.561 over the same 15 patients. This agrees with the weak direction decoding reported here.
supportsMovement-detection channels did not improve direction decoding over random channels in 15 SEEG patients
Our all-channel direction AUC (0.571) agrees with the earlier all-channel result (0.567) in the same 15 patients.
supportsIn 15 ds006253 SEEG patients, three 0.2 s HGA bins did not improve offline direction decoding over their mean; bin shape was at chance
With the same 15 patients and decoder family, our 0-0.6 s window-mean direction AUC (0.574, 0-1.0 s: 0.581) agrees with its weak above-chance direction decoding (mean AUC 0.567). We add that time-resolved 0.2 s bins do not improve it and that bin shape alone is at chance.
supportsFive stacked 0.2 s bins did not improve left/right SEEG direction decoding over the 0–1 s mean in 15 patients
With the same patients and decoder, ST direction AUC was again weakly above chance (0.569, 10/15) and peaked at 0.4-0.6 s. We extend the test to the 0-1 s time course and find no gain over the window mean.