Multimodality white matter tract segmentation using CNN

Diffusion-weighted MRI tractography combining with fiber clustering is a typical approaches of white matter tract segmentation. However, due to the anatomical accuracy of the diffusion-weighted images (DWI) and complicated processing procedure, this approach suffer many inherent limitations that it is difficult to handle complex fibers well and might introduce subtle errors in further processing steps. In this paper, we proposed a novel white matter tract segmentation method with dual input convolutional neural network to fuse two modal data. Specifically, the structural tensor of T1-Weighted (T1W) MR images was introduced to complement to the peaks of fiber orientation distribution function (fODF). The proposed method has been experimented with 105 subjects of Human Connect project and proved to more accurately identify white matter bundles than existing methods.