Under assumed timing, SEEG load readers showed no clear AUC difference between hippocampus and temporal cortex in 12 final sessions

We compared small SEEG readers from hippocampus and temporal cortex. We used public OpenNeuro ds004752 version 1.0.1. Both contacts of each pair needed a label in the target region. Twelve patients met this rule. Each reader used the same one or two pair budget. Earlier sessions selected the pairs and trained each reader. Final sessions supplied 338 test trials. The primary signal was log power at epoch seconds 4 to 5 minus log power at seconds 0 to 1. Phase names require the assumption that each epoch starts at fixation. Both readers also used task history. Mean AUC was 0.556 for hippocampus and 0.583 for temporal cortex. The paired difference was 0.027. Its 95 percent patient bootstrap interval was -0.069 to 0.128. Six patients favored each region. This result does not establish equal performance. A planned secondary check with raw power favored temporal cortex, but it needs a separate test. We checked 72 fresh source windows and retained them. We also changed all final-session signals and load labels; all 180 fitted models stayed unchanged. The package includes figures, code, source records, features, predictions, and rebuild instructions. This is an exploratory analysis of reused data. It does not test online control or stimulation.

stereo-EEG · verbal working memory · letter load · hippocampus · temporal cortex · area under the curve · cross-session decoding · exploratory analysis