It has been recently reported that the spatio-temporal correlation of white matter BOLD signals in resting-state functional MRI (rs-fMRI) can be captured using functional correlation tensors (FCTs). FCTs exhibit anisotropy information similar to diffusion tensor imaging (DTI). In this work, we employ a patch-based strategy to improve the noise-robustness of FCTs. Then, we adopt regression forest to learn a mapping from FCTs to DTs. Testing using unseen images, the predicted DTs show high similarity with the actual DTs. This validates the fact that FCTs carries information that is highly correlated with DTs.