Previous-trial load was not decoded from 3–6 s SEEG band power or current-load scores in 15 patients

Question: Does the memory load of trial i−1 remain in the SEEG band power of trial i and bias a load decoder? Data: OpenNeuro ds004752 verbal working-memory task, 15 patients, session 1, with archived log band power from five bands and bipolar channels in the 3–6 s maintenance window. A shrinkage LDA with stratified cross-validation decoded current load (AUC 0.70, 95% CI 0.62–0.78, p=0.0007). After removing current set size, previous load was not decoded (AUC 0.530 vs permutation-null mean 0.503; difference CI −0.04 to +0.09; p=0.44; 1/15 patients p<0.05). Previous load did not shift the current-load decoder score (−0.03 SD, CI −0.33 to +0.27, p=0.86), and repeat- and switch-trial accuracy did not differ (0.59 vs 0.66, p=0.42). For 8-s trials, 3–6 s band power, and this electrode coverage, group data exclude carry-over larger than about 0.09 AUC above null or 0.3 SD in decoder shift. The study does not test shorter trials, other signal features or windows, or small patient-level effects; it does not show that no brain region holds trial history.

SEEG · verbal working memory · trial history · memory load · band power · decoding · shrinkage LDA · brain-computer interface