CNLNet: Enhancing MRI Brain Image Denoising Using a Convolutional Neural Network with Integrated Non-Local Means Layer

Abstract

Magnetic resonance imaging (MRI) is an essential imaging modality for brain, but it is affected by noise which degrades the quality of the images. Traditional methods often fail to maintain the fine details of the image while providing denoising efficiency. Even though deep learning methods perform well, integrating them with traditional approaches for medical image denoising is under explored. In this work, we developed CNLNet, Deep learning(DL) model that combines convolutional neural network (CNN) with Non-local means (NLM) layer for MR brain image denoising. Local feature extraction is done by utilising CNN. NLM layer refines long-range dependencias, delivering better denoising performance. Model was trained on Kaggle MRI brain dataset with varying Gaussian noise of standard deviation 0.1, 0.3, 0.5. Quantitative results of experimentation reveal that CNL Net performs well when compared to the CNN model, traditional median, Wiener, NLM filters in terms of PSNR, SSIM, MSE, MAE. Visual comparison also highlights that CNLNet preserves significant fine details along with noise removal. This approach improves the quality of MRI brain images demonstrating potential for clinical diagnostics applications and offers a more efficient denoising solution compared to conventional methods.