Distortion Removal and Deblurring of Single-Shot DWI MRI Scans

Ahana Roy Choudhury, Sachin R. Jambawalikar, Piyush Kumar, Venkat Sumanth Reddy Bommireddy

Abstract: Diffusion Weighted Imaging (DWI) is one of the standard MRI images that are used for the diagnosis of brain tumors. However, the acquired DW images suffer from artifacts such as EPI (echo-planar imaging) distortion. These distortions are corrected using blip-up and blip-down images that are separately acquired for distortion removal. In addition, DWI MRI images have shorter acquisition times, but suffer from poor resolution. Multi-shot diffusion weighted imaging allows the acquisition of higher resolution images but require longer acquisition times and necessitate the use of new and expensive hardware. In this paper, we perform distortion removal of EPI-DWI images from blip-up images using a previously proposed framework and design a suitable deblurring technique for generating higher resolution DWI images from low-resolution undistorted EPI-DWI images. Our technique aims to allow the use of deblurred EPI-DWI images for performing accurate medical diagnosis and multi parametric longitudinal analysis in brain tumors. We use data augmentation, dilated convolution, and ELU (exponential linear unit) to design a suitable architecture that achieves superior performance in terms of accuracy.