Super Resolution for Magnetic Resonance Images Using Self-Super Resolution Technique
P. T. Vasanth Raj, A. Vijayaraj, R. Dhanagopal, R. Suresh Kumar, M. Ayyadurai · 2022 6th International Conference on Devices, Circuits and Systems (ICDCS) · 2022
Recently, the need for MRI images in clinical applications has skyrocketed. While this is very desired, a trade-off will exist between acquisition speed, resolution, and noise. By and large, MRI pictures have a higher in-plane resolution than through-plane images. It is impossible to get high frequency information from these pictures through through-plane and it cannot be derived by interpolation either. To address this issue, the super resolution (SR) approach was created to improve spatial resolution. SR approaches learn the transition from lowresolution (LR) to high-resolution (HR) pictures using a training set. HR atlas pictures are not readily accessible for MRI for a variety of reasons. In this article, we offer a self-SR method that is not reliant on an external training data. However, it is possible to bypass this limitation using HR pictures, which are dependant upon obtaining LR images. To create training sets, several forms of distorted input pictures are employed. To estimate the HR picture, the trained set is applied to the input image. When compared to existing self-super-resolution (SSR) approaches, the suggested result improves through-plane resolution significantly.