Fig. 1:

Schematic of CAROT: A) During training, CAROT transforms time series fMRI data from multiple source atlases to a target atlas to obtain transportation mappings. Mappings between the source and target atlases are found by employing optimal transport and solving Monge–Kantorovich transportation problem using the Sinkhorn approximation. The solution provides a transformation that maps the brain activity parcellated using the source atlas to brain activity parcellated based on the target atlas. B) During testing, for each pair of source and target atlases and a single time point in the time series data, the offline solutions are used, and time series and functional connectomes accordingly will be reconstructed in the desired target atlas. Results from several pairs of source and target atlases can be combined to improve the quality of the final reconstructed connectome. C) A standard image preprocessing pipeline to create functional connectomes.