In 13 sessions, evoked SEEG voltage predicted working-memory errors less well than band power, while induced and total band power performed similarly

Question: does the stimulus-locked (evoked) SEEG voltage contain information about later working-memory errors that band power does not contain? We used 13 sessions (12 patients) of OpenNeuro ds006136 (load-3 object working memory, middle frontal and middle temporal contacts). We used the same per-channel shrinkage-LDA ensemble and 5-fold cross-validation for each feature set. Band power gave median AUC 0.651 (13 of 13 sessions above 0.5). Evoked voltage bins gave median AUC 0.519 (8 of 13, p = 0.073) and were lower than band power in 11 of 13 sessions (p = 0.003). Evoked plus band power did not do better than band power alone (median difference -0.011, p = 0.17). Band power after removal of the mean evoked response gave the same AUC (0.648, p = 0.84). A positive control showed clear evoked responses (12% to 67% of channels with |t| > 4). Thus the error information is in induced band power, not in the phase-locked voltage. Limits: offline, within-session decoding, 5-18 cortical contacts, n = 13, one classifier and one evoked feature design. The package has code, per-session results, figures, data hashes and reproduction steps.

SEEG · working memory · error prediction · evoked voltage · band power · induced activity · AUC · within-session decoding