Abstract:Magnetic resonance imaging (MRI) has been applied in various fields, especially for the medical purposes. However, fine details and smooth areas of some critical patterns in an MR image polluted by common thermal noise will interfere with the diagnosis of doctors. Thermal noise in an MR image obeys Rician distribution, which is hard for conventional denoising methods based on shift invariant spatial filtering approaches to dispose. Besides, the fine detail and edge information will be inevitably damaged when smoothing the noise, which is unacceptable for medical images. In this paper, we propose two corresponding solutions. First, we design a convolutional neural network (CNN) to learn a mask that aims to directly eliminate the thermal noise in the background region, and make the noise in the MR image obey almost the same distribution. Second, we propose several improvements in terms of existing deep learning approaches for thermal noise removal. Specifically, we establish a dual-branch neural network, a frequency-domain-optimizable discrete cosine transform (DCT) module, and adopt other effective structures such as residue blocks, convolutional block attention modules (CBAM), and parallel multi-dilated GoogLeNet inception based convolutional blocks to form an attentive residue multi-dilated network (ARM-Net). We evaluate our method over the BraTS 2018 dataset at noise levels ranging from 2% to 20%. Experimental results reveal that our method achieves state-of-the-art (SOTA) performance compared with the most recent works.

Edge Preserving Filtering by Combining Filters for Magnetic Resonance Image

High Quality Image-Pair-based Deblurring Method Using Edge Mask and Improved Residual Deconvolution

Non-quadratic regularized edge-preserving reconstruction for parallel magnetic resonance image

Remove Impulsive Noise For Infrared Image With Difference Images And Adaptive Filter

Ultrasound Image Edge Detection Based on a Novel Multiplicative Gradient and Canny Operator.

Effective hybridization approach for noise removal in magnetic resonance imaging

Eliminating Noise of MR Image with Wavelet Transform

Denoise ultra-low-field 3D magnetic resonance images using a joint signal-image domain filter

A Multiscale Gradient Fusion Method for Edge Detection in Color Images Utilizing the CBM3D Filter

Study of MRI Image Denosing Method Based on MCA and Context Model

An Adaptive Anisotropic Diffusion Filter for Noise Reduction in MR Images

An Image Filter Based on Multiobjective Genetic Algorithm and Shearlet Transformation

A cardiac MRI motion artifact reduction method based on edge enhancement network

An improved adaptive image filter for edge and detail information preservation

An Improved Attentive Residue Multi-Dilated Network for Thermal Noise Removal in Magnetic Resonance Images

Edge Detection Algorithm Based on Self-adaptive Filter for Noise Image

Guided Frequency Filter for Block-DCT Compressed Capsule Endoscopic Images

A Novel Edge Protective Adaptive Filter for High Energy X-ray Imaging Technology

An Information-Fusion Edge Preserving Method in Image Filtering

A daptive edge detection algorithm of MRI image based on morphology

Mr Image De-Noising Algorithm Based on Improved Weickert Structure Tensor Model