GAN-CNN based Structure-Preserving Mixed Noise Removal Model for Enhancing Medical Image

GAN-CNN BASED STRUCTURE-PRESERVING MIXED NOISE REMOVAL MODEL

Abstract

The current era of the Internet of Medical Things (IoMT) and Medical Artificial Intelligence (MAI) makes medical imaging a prominent mode of providing effective solutions in diagnosis and prognosis. The main issue with these images is the presence of noise that requires enhancement through effective edge preservation and noise reduction. The proposed work introduces a two-stage Deep Learning (DL) model, utilizing Generative Adversarial Networks (GANs) and Convolutional Neural Networks (CNN) for jointly reducing speckle, impulse, and Gaussian noise while preserving edge information in noisy medical images. The work also explores the probabilistic evaluation of generators and discriminators for compensating lossy patches to ensure image quality. The performance of the proposed model is investigated by considering three different performance metrics, namely, PSNR, FSIM, and SSIM. Moreover, non-parametric statistical tests like the Sign test, Wilcoxon Signed rank tests and Friedman tests are also conducted to assess the dominance of the proposed model over other state-of-the-art approaches. Two-stage GAN-based models generate realistic, high-quality images by effectively suppressing inherently present spurious noise in medical images and simultaneously preserving the edge information.