Spread SEEG channel placement improved memory decoding in three computable tests

We tested whether spreading a fixed number of stereo-electroencephalography (SEEG) bipolar pairs across shafts improved memory decoding compared with placing adjacent pairs on one shaft. We analyzed archived features for 19 subjects in OpenNeuro ds004789 (free recall) and 11 subjects in ds004752 (Sternberg working memory). For each eligible subject and channel budget, we compared 100 random dense sets with 100 random spread sets using shrinkage linear discriminant analysis, high-gamma and beta power, and five-fold cross-validation. Spread placement increased mean area under the receiver operating characteristic curve (AUC) in all three computable tests: FR1 K=4, +0.018 (15/19 subjects; Holm-adjusted p=0.0019); FR1 K=8, +0.019 (14/19; p=0.0086); and working memory K=4, +0.029 (10/11; p=0.024). The preregistered working-memory K=8 test could not be computed because no shaft had eight bipolar pairs. Exploratory K=7 gains were +0.019 for FR1 (p=0.0057) and +0.036 for working memory (p=0.0625); these tests were outside the Holm family. The memory-decoding gain at K=8 was about 35% of the reported motor-decoding gain. This analysis tested random placement only, used one decoder and two band-power features per pair, and did not distinguish effects of brain-region coverage from between-shaft noise correlation.

stereo-electroencephalography · SEEG · memory decoding · free recall · working memory · channel placement · bipolar pairs · shrinkage LDA · AUC · high-gamma power · beta power