In 10 Patients, Equal-Count SEEG Decoding Found No Significant Difference Between Temporal and Frontal Contacts for Working-Memory Errors

We tested whether middle temporal gyrus (MTG) or middle frontal gyrus (MFG) bipolar SEEG contacts better predicted later memory errors in a load-3 object working-memory task. We used archived log10 band-power features from OpenNeuro ds006136 version 1.0.1 and a cross-subject L2 logistic decoder. The analysis plan was written before the regional comparison. In 10 patients with contacts in both regions, equalizing the number of channels gave a mean AUC difference (MTG minus MFG) of +0.025 (95% CI -0.032 to +0.088; exact two-sided sign-flip p = 0.49). MTG contacts predicted errors above chance (median AUC 0.594, p = 0.006), while MFG contacts did not (median AUC 0.548, p = 0.09); this difference in significance does not establish a difference between regions. Adding MFG contacts to MTG contacts produced a median AUC change of -0.002. Region-specific models did not outperform a pooled model. The pooled model reproduced the source result (median AUC 0.590; 10 of 12 patients above chance). The evidence supports temporal contacts as sufficient for cross-patient error decoding in this task, but does not show that frontal contacts are necessary or better. Limits include 10 paired patients, clinical electrode placement, small effects (AUC about 0.55 to 0.60), one task, and two regions; this is not yet a usable single-trial BCI signal.

working memory · SEEG · brain-computer interface · cross-subject decoding · middle temporal gyrus · middle frontal gyrus · band power · memory errors