Single Image Based Reconstruction of High Field-Like MR Images

Abstract

Unavailability of high field magnetic resonance (MR) scanners has encouraged the development of post-processing algorithms to reconstruct 7T-like MR images from the given 3T MR images. Such algorithms have recently been developed with promising performance, but with the expensive requirement of paired 3T-7T MR images. Due to practical unsuitability to acquire example paired 3T-7T images in the real scenario, in this paper we explore the potential of the single image-based approach to reconstruct the 7T-like MR images. The intensity transformation between 3 T and 7T MR images is assumed to be linear and is initialized as Toeplitz form of Gaussian kernel in alternative minimization framework to reconstruct 7T-like images from 3T MR images. Further, to achieve the desired contrast we compute the relation between statistics of the intensity values of two different tissues for each of the original 3T & 7T and registered 3T MR image. This relation is used to constrain the solution space for 7T-like MR image reconstruction. The proposed algorithm provides comparable performance to the existing algorithms which require 3T-7T images pair. The qualitative and quantitative analysis is done for reconstruction and segmentation results, indicating the advantages of proposed work.