Optimal Diffusion Tensor Restoration

Abstract

Assessing the effects of white matter (WM) lesions on structural connectivity as measured by diffusion MRI (dMRI) is invaluable for understanding structure-function relationships. These WM lesions have many etiologies that ultimately lead to attenuation of the anisotropic signature in dMRI signals. Attenuation can produce inaccurate reconstructions of the underlying diffusion tensors (DTs). In this paper, we combine methods from image inpainting and estimation theory to develop a novel approach for restoring DTs in small and moderate sized WM lesions. Our approach begins by taking healthy WM DTs at the boundary of the lesion and filling in lesioned voxels with their optimal affine estimate moving iteratively in a fast-marching method style until the DTs in the lesion are restored. We demonstrate with in-vivo simulations that our approach offers superior performance over a baseline inpainting approach. We restore lesions in three stroke patients suffering hemiparesis from damaged corticospinal tracts (CST). We show that our method restores anisotropy and orientation of lesioned tensors enabling reconstruction of the corticospinal tract using tractography.