Enhancing Diffusion MR Tractography Using a Deep Learning Model that Incorporates Anatomical Knowledge

Motivation: To develop reliable diffusion MRI tractography to study brain connectivity. Goal(s): The study aims to improve the estimation of fiber orientation distribution (FOD), which is key to improve the specificity of tractography. Approach: We created an augmented streamline dataset based on known white matter pathways to train a deep neural network to estimate FOD from diffusion MRI signals. Results: Tractography based on the network estimated FODs showed reduced false-positives compared to conventional methods. The improvement remained for input data with reduced angular resolutions and added noise. Impact: The proposed method can improve tractography by reducing false-positives and benefit studies on structural connectivity of the brain. Furthermore, it may shorten the acquisition time required for robust tractography, which is important for studies on children and seniors.