Figure 2.

Single category feature decoding with averaged trials. (A) A schematic view of analysis procedure. The decoders trained to decode the DNN feature values calculated from the presented images in the training dataset were applied to an averaged fMRI activity associated with a specific dream category to test decodability of the values of the category features constructed from multiple images annotated by the dream categories. (B) Correlation coefficients between the category feature values and the decoded feature values [error bars indicate 95% CI across feature units; two subjects pooled; three-volume (9 s) averaged fMRI signals immediately before awakening].