Complete Oriented Deviation Texture Loss Based GAN (CODT_GAN) for MRI Image Denoising

Abstract

The use of denoising is crucial in MRI applications for diagnosing medical conditions. MRI pictures frequently include undesirednoises that might impair the accuracy of a pathological diagnosis. Deep learning networks have recently been used to generate a large number of models for MRI denoising with excellent results. However, the majority of them are incapable of providing clients with additional perceptual features of the image without information loss. In this paper, a novel structural and texture loss approach named Complete Oriented Deviated Texture Loss (CODTL) based GAN (CODT_GAN) was proposed to complete the image denoising task. Initially, a GAN network was used to produce images to resemble original images as closely as feasible. Then the GAN generated image is fed into the proposed CODTL operator to estimate textural loss. This proposed approach incorporates structural loss and textural loss, which has been used to fill in the gaps left by GAN network denoising of MRI images. The effectiveness of the proposed work is evaluated by BraTS’17 dataset. The results of the experiments show that the suggested approach is more reliable and performs better than the traditional methods in terms of both the degree of denoising and retention of the anatomical structures, textural details, and defined contrast.