The 20-45 Hz aperiodic SEEG slope did not track verbal working-memory load in four patients

We tested whether the 20-45 Hz aperiodic slope of stereo-electroencephalography (SEEG) during maintenance changes with verbal working-memory load, and whether aperiodic features decode set size 8 versus 4 as well as band power. We analyzed session 1, run 1 from four patients in OpenNeuro ds004752 v1.0.1: 119 bipolar channels and artifact-free trials from a 3-6 s maintenance window. A fixed shrinkage linear discriminant decoder used repeated stratified 5-fold cross-validation and 200 label permutations. The 20-45 Hz slope showed no consistent load relationship (mean channel Spearman rho +0.002; no channel survived FDR correction) and did not decode load above chance in any patient (mean AUC 0.62; Fisher-combined p=0.09). Band power had mean AUC 0.66 (Fisher p=0.006), although only one patient passed individually. A broad 3-45 Hz fit had mean AUC 0.68, but includes oscillatory peaks and is not evidence that the aperiodic component carries load information. Adding aperiodic features changed mean AUC from 0.66 to 0.69, with inconsistent patient-level changes. Results are limited by four patients, one session each, few trials, a simple fit without peak removal, and no anatomical channel selection. These data do not support the 20-45 Hz slope as a useful single-trial read-out in this sample; effects may have been missed because of low power.

stereo-electroencephalography · working memory · aperiodic slope · band power · memory load · brain-computer interface · decoding · OpenNeuro