Slow mouse movements had higher SEEG detector miss rates in 6 of 7 patients; trial-level score-vigor correlation was not significant

We tested whether a 70–150 Hz SEEG move-vs-rest detector missed slow movements more often than fast movements. We analyzed archived author HGA features from OpenNeuro ds006253 v1.0.3 in 7 patients (50–97 bipolar channels; 187–300 trials each). Rest was the pre-stimulus window; movement was 0–0.5 s after movement onset; vigor was peak mouse velocity. After excluding 3 trials with vmax < 1, shrinkage LDA was trained on the first 60% of time-ordered trials and tested on the last 40%. The pre-registered primary outcome, partial Spearman correlation between test movement-window score and log peak velocity (adjusted for trial index and movement latency), averaged +0.105 (4/7 positive; exact one-sided p=0.086), not significant. The slowest-third miss rate exceeded the fastest-third rate by 0.135 (6/7 positive; p=0.016; Holm p=0.047), with differences of +0.20 to +0.35 in 3 patients. Two patients had miss rates near 0.9 in both thirds. The slow-minus-fast AUC difference was -0.072 (5/7 negative; Holm p=0.125), not significant. Rest-window score was not associated with vigor (rho=+0.016, p=0.78). Mean test AUC was 0.78 and mean false-positive rate was 0.093 (target 0.10). Exploratory channel analysis found mean rho +0.006 and |rho| > 0.2 in 3.5% of 566 channels. An exploratory fast-only training result is not interpretable because its threshold used in-sample rest scores. With 7 patients, H1 power is low; a null H1 does not show no effect. Peak mouse velocity during hand movement is not BCI attempt strength. Only one post-onset window and one decoder were tested.

SEEG · brain-computer interface · movement detection · movement vigor · high-gamma activity · linear discriminant analysis · OpenNeuro