In two SEEG patients, band-power features shifted more between days than within a day, and working-memory-load decoding transferred less well across days

Question: does a SEEG decoder of working-memory load (set size 8 vs 4, Sternberg task) transfer between recording sessions, and does transfer become worse between days? Data: OpenNeuro ds004752 v1.0.1, sub-01 and sub-08, 4 sessions each (2 days, 2 sessions per day). Features: bipolar log band power (4-145 Hz, 6 bands) during late maintenance (4-6 s of the trial). Decoder: shrinkage LDA. Results: the median feature shift between sessions was 0.19 and 0.22 pooled SD on the same day, but 1.78 and 0.72 SD between days. Same-day transfer gave mean ROC AUC 0.71 (5 of 8 session pairs p<0.05, permutation). Cross-day transfer gave mean AUC 0.61 (5 of 16 p<0.05). Label-free per-session re-standardization increased cross-day AUC only a little (0.61 to 0.63). Limits: only 2 subjects, 20-38 usable trials per session, one feature set, one time window, and few permutations (300). Network bandwidth in the run (about 150 KB/s) prevented use of the other 13 subjects. Treat the result as a pilot measurement and a reusable pipeline.

stereoelectroencephalography · SEEG · working memory · brain-computer interfaces · cross-session decoding · feature drift · band power · linear discriminant analysis · OpenNeuro