Time-bin features did not improve left/right movement decoding in 15 SEEG patients

In a preregistered offline analysis of 15 patients with stereoelectroencephalography (SEEG), we tested whether five 0.2 s bins of 70–150 Hz activity from 0–1.0 s after movement onset improved left/right direction decoding over one 0–0.5 s window mean. With correct trials, Ledoit–Wolf shrinkage linear discriminant analysis, and five contiguous time-ordered folds, the mean AUC difference was −0.002 (95% bootstrap CI −0.033 to +0.030); five-bin features were better in 6/15 patients (one-sided exact sign-flip p = 0.54). We found no evidence of improvement. Against a single 0–1.0 s mean, the five-bin features were slightly worse (mean difference −0.019, 95% CI −0.037 to −0.001; 3/15 better; one-sided p = 0.97, Holm p = 0.97; descriptive direction only). The five-bin features were above chance (mean AUC 0.571, 12/15 patients; p = 0.0024, Holm p = 0.0048), but direction decoding remained weak across feature sets (mean AUC 0.57–0.59). The exploratory 0–1.0 s mean comparison with the 0–0.5 s mean was +0.017 (9/15 better, p = 0.058). For this linear decoder and trial count, a single window mean is a reasonable starting feature. This offline study used one session per patient and windows aligned to known movement onset; it does not rule out gains from nonlinear decoders, more trials, a lower-dimensional time basis, or other frequency bands.

stereoelectroencephalography · brain-computer interface · movement direction · high gamma activity · linear discriminant analysis · time course · decoding