Fig. 2:

Using multiple source atlases improves the similarity of reconstructed connectomes. A) The Spearman’s rank correlation between the reconstructed connectomes and connectomes generated directly with the target atlases are shown for each pair of source and target atlas as well as reconstructed connectomes using all of the source atlases. Using all source atlases produces higher-quality reconstructed connectomes for each target atlas. Error bars are generated from 100 iterations of randomly splitting the data into 25% for training and 75% for testing. B) For each target atlas, increasing the source atlases increases the similarity of reconstructed and original connectomes. For most atlases, a Spearman’s correlation of $\rho >0.60$ (red line) can be achieved by using fewer than five source atlases (i.e., all available source atlases). Circle size represents the variability of the correlation over 100 iterations of splitting the data into training and testing sets.