Sheng Chen, The University of Sydney
Diffusion-weighted MRI (dMRI) is an invaluable MRI technique in neuroimaging analysis, since it enables non-invasive probing of brain microstructures. A diffusion tensor imaging (DTI) model can be fitted on diffusion-weighted imaging (DWI) to characterize brain tissues by extracting several diffusion measure metrics such as Fractional Anisotropy (FA), Mean Diffusivity (MD), Axial Diffusivity (AD), and Radial Diffusivity (RD). These metrics are widely used in clinical studies for observing the group differences between health controls and patients. However, clinical acquisition constraints often lead to low angular resolution diffusion imaging (LARDI) and diffusion measure metrics derived from LARDI are unreliable. To obtain trustworthy diffusion measure metrics from LARDI for clinical studies, we propose High Angular Resolution Diffusion Tensor Imaging Estimation Network (HADTI-Net) to generate the enhanced low angular resolution DTI (LAR-DTI) from a minimal set of evenly distributed diffusion-weighted directions. We trained and evaluated HADTI-Net on the Human Connectome Project (HCP) dataset. The results show that the enhanced LAR-DTI by HADTI-Net was able to derive diffusion measure metrics comparable to those derived from high angular resolution DTI (HAR-DTI). Further extensive experiments demonstrate HADTI-Net’s clinical impact in reducing diffusion measure metric differences, enabling the observation of group distinctions between healthy controls and patients with neurological diseases.