Matched-count MTG and MFG SEEG channels did not differ significantly in working-memory error prediction

This pre-registered comparison asked whether middle temporal gyrus (MTG) or middle frontal gyrus (MFG) stereo-EEG channels better predicted later recall errors when channel counts were matched. The analysis used archived per-channel log10 Welch band-power features from the load-3 object working-memory task in OpenNeuro ds006136 v1.0.1. For each of 10 subjects with channels in both regions, per-channel shrinkage-LDA decision values were combined in an ensemble and evaluated with stratified five-fold cross-validation over five seeds. Twenty random subsets of the same matched channel count were compared per region and seed. The median AUC difference (MTG minus MFG) was -0.044; MFG was higher in six of 10 subjects (exact two-sided Wilcoxon p = 0.160; subject differences ranged from -0.162 to +0.066). The data did not show a significant region difference, which does not establish equivalence. At matched counts, MTG AUC was above chance in 9/10 subjects (median 0.542, one-sided p = 0.0049) and MFG AUC in 10/10 (median 0.618, p = 0.00098). The sample was small, channel counts were 1–9 (median 4), electrode placement was clinical, and region and hemisphere were partly confounded. These limits and the low power for a median difference of about 0.04 AUC restrict interpretation.

stereo-EEG · working memory · error prediction · middle temporal gyrus · middle frontal gyrus · shrinkage LDA · cross-validation · AUC · brain-computer interface