Single-channel SEEG predicted working-memory errors above chance in 12 patients, but the LMFG–RMTG advantage was not significant after correction

We tested whether single stereoelectroencephalography (SEEG) channels in the left middle frontal gyrus (LMFG) and right middle temporal gyrus (RMTG) predict a later working-memory (WM) error better than channels in the right middle frontal gyrus (RMFG) and left middle temporal gyrus (LMTG), as a decoding consequence of the source paper’s report that LMFG and RMTG support WM. We analyzed 138 channels in 12 patients from OpenNeuro ds006136, using log band power (4–150 Hz, five bands) in encoding and delay windows, shrinkage linear discriminant analysis, and stratified five-fold cross-validation with five repeats. Across all channels, the mean of patient means was AUC 0.553; 10 of 12 patients had AUC above 0.5 (sign-flip p = 0.0034). LMFG+RMTG exceeded RMFG+LMTG by 0.034 AUC in 7 of 9 patients (exact sign-flip p = 0.047; within-patient label-permutation p = 0.056). This p-value was not corrected for five contrasts; after Holm correction, no contrast was significant. Left minus right was +0.008 (p = 0.81), and frontal minus temporal was +0.037 (p = 0.082). The small LMFG+RMTG advantage cannot be separated from a frontal-over-temporal effect, because LMTG had the lowest regional AUC and LMFG and RMFG were both sampled in only one patient. These results do not support a strong lateralized difference in error read-out. The analysis was offline, within-session, and single-channel; it used processed bipolar trials after artifact rejection by the dataset authors, 12 patients, one feature set, and one classifier.

working memory · stereoelectroencephalography · error decoding · single-channel classification · linear discriminant analysis · cross-validation · OpenNeuro