Fig. 3: Highest cross-regional generalization in the EC.

A Schematic of cross-regional decoding analysis. Using the entorhinal cortex (EC) as an example, we trained classifiers using power features from EC for each trial and predicted the load using power from the hippocampus (Hipp) for both low-to-medium and medium-to-high load conditions. The specific decoding steps were the same as shown in Fig. 2C. For all brain regions, models were trained using their own power features and tested on data from the other two brain regions. The generalization of each brain region was determined by averaging its accuracy when tested on data from the other two brain regions (hippocampus: light red; EC: light blue; lateral temporal cortex: LTC, light green). The brain was visualized by the BrainNet Viewer toolbox (www.nitrc.org/projects/bnv/)⁴³. B Accuracy matrix of cross-regional decoding on low-to-medium load (left) and medium-to-high load (right). The rows of the matrix represented the regions used for training, while the columns denoted the regions employed for testing, with the values representing the average accuracy. C The averaged cross-regional decoding accuracy for load 4 vs load 6 did not differ significantly among hippocampus, EC, and LTC across all cross-validations (n = 100; two-sided permutation t tests: EC vs hippocampus: p = 0.637; EC vs LTC: p = 0.128; LTC vs hippocampus: p = 0.141). The EC showed the highest cross-regional decoding accuracy for load 6 vs load 8 across all cross-validations (n = 100; two-sided permutation t tests: EC vs hippocampus: p < 0.001; EC vs LTC: p = 0.002; LTC vs hippocampus: p < 0.001). **p < 0.01, ***p < 0.001. The center line represents the median, and the edges of the box correspond to the lower and upper quartiles, respectively. The whiskers extend to the minimum and maximum data points at most 1.5 times the interquartile range. Source data are provided as a Source Data file.