In 8 new ds006253 patients, a 70–150 Hz detector missed slow mouse movements more often and its score rose with peak velocity

We tested whether a high-gamma SEEG movement detector misses low-vigor movements more often in the eight ds006253 patients not used in sci 6fe90d33. We used archived author 70–150 Hz HGA features and the unchanged detector code, with a pre-registered plan. A Ledoit–Wolf shrinkage LDA used log10 HGA from all bipolar channels, movement at 0–0.5 s after onset, and rest in the 0.5 s before the stimulus. The first 60% of trials trained the detector and the last 40% tested it; its threshold was the 90th percentile of out-of-fold training rest scores. In the eight new patients, the slowest-third minus fastest-third miss-rate difference was +0.076 (95% CI +0.021 to +0.131; 6/8 positive; Holm p = 0.023). Partial Spearman correlation between test movement score and log peak velocity, controlling for trial index and latency, was +0.131 (95% CI +0.060 to +0.192; 7/8 positive; Holm p = 0.016). The rest-score control was rho +0.016 (p = 0.68), so the association was not a trial-level offset. Mean test AUC was 0.874 and the test false-positive rate was 0.096. In the original seven patients, the miss-rate difference was +0.135 (6/7; Holm p = 0.031), while the trial-level correlation was not significant (+0.105; 4/7; Holm p = 0.086). Across all 15 patients, the corresponding values were +0.104 (12/15; Holm p = 0.0023) and +0.119 (11/15; Holm p = 0.0023). An exploratory decoder trained only on high-vigor trials reduced the slow-third miss rate from 0.373 to 0.242, but its threshold used in-sample rest scores and is optimistic. This was an offline analysis of one mouse-decision task, one session per patient, and one 0–0.5 s window. Hand-movement vigor in patients who can move is not the strength of an attempt in a paralysed BCI user. Miss rates were near ceiling in some patients, test-set tertiles contained 48–120 trials, and the average effect was small to moderate (7.6 percentage points).

stereo-EEG · high-gamma activity · movement detection · movement vigor · brain-computer interface · shrinkage LDA · replication