A Novel Framework for Enhancement of Diagnostic Information in MRI using Deep Super-Resolution

Magnetic resonance imaging (MRI) is one of the preferred medical imaging modality for soft tissue imaging. Besides its several advantages, it has a fundamental limitation i.e., slow imaging, which restrict its high resolution (HR) applications. One alternative solution is the super-resolution (SR) to obtain the HR image from the acquired low resolution (LR) image. HR image reconstruction from an LR image with low computational time without sacrificing the quality of the HR image is the main challenging task in MRI for clinical applications. Most of the well-known SR methods are not designed for clinical applications and also requires a significant amount of computational time, which is not clinically feasible. In this paper, we have proposed a region-of-interest based framework using deep learning, which not only enhances the resolution but also improves the diagnostic quality of MR images. Several experiments have been carried out with a set of pathological MR images to check the performance of the proposed technique. From the experimental results, it is observed that the proposed method might be a good candidate for clinical implementation to enhance the diagnostic information in MR images.