Movement related high gamma decreases add held out information in 13 SEEG subjects

Archived trial features from 15 subjects in two datasets were used to test whether bipolar stereoelectroencephalography (SEEG) pairs with a movement related decrease in 70–150 Hz high gamma (DEC) add held out movement information to pairs with an increase (INC). Pair labels were set from training trials only using a paired t test of log power (two sided p < 0.01). In the pre-registered primary test, shrinkage linear discriminant analysis (LDA) using INC and DEC pairs had higher time later held out AUC than INC pairs alone in 13 subjects with both pair types (0.785 vs 0.724; mean difference +0.061; all 13 positive; exact two sided sign flip p = 0.0002). DEC pairs alone detected movement above chance (mean AUC 0.667; all 13 above 0.5). A fixed sign decoder also gained from subtracting DEC pairs (+0.044; 10 of 13; Holm adjusted p = 0.033). An exploratory control found a greater gain than adding the same number of random non-significant pairs (+0.045; 12 of 13; p = 0.002). DEC pairs were about as frequent as INC pairs (112 vs 107 of 1393). These offline results cover 15 subjects, two tasks, one frequency band, one decoder and one time ordered split per subject. A bipolar decrease may reflect higher correlation between contacts rather than lower local activity. The ds006253 rest window was pre-stimulus fixation, and the ds004473 rest window was task rest. The small k-matched test (three subjects) was not informative. The package preserves the pre-analysis plan, code, archived features and their hashes, results and figure.

stereoelectroencephalography · high gamma · movement decoding · bipolar pairs · brain computer interfaces · linear discriminant analysis