Complex-Domain Deep Learning for MR Image Denoising
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing deep learning-based MR image denoising methods fail to effectively address Rician noise bias, leading to biased denoised images and potential loss of small anatomical features due to over-smoothing.
Innovation Solution
A deep learning-based method that processes the real and imaginary parts of MR images using a complex-valued convolutional neural network, splitting frequency components to preserve low-frequency information and control noise reduction, with adjustable variance and bias settings to optimize denoising strength.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Object-affected harmful factors
If conventional filtering techniques or deep learning-based methods are used to denoise magnitude MR images, then noise is reduced, but Rician noise bias is introduced and cannot be completely removed
Solution Approach 1:
Instead of denoising the magnitude image directly (conventional approach), the invention inverts the processing order by denoising the complex-valued image first (real and imaginary parts separately), then computing the magnitude image. This inversion avoids the Rician noise bias that plagues direct magnitude image denoising.
Solution Approach 2:
The invention changes the parameter space from magnitude images to complex-valued images. By operating in the complex domain where noise follows a Gaussian distribution rather than a Rician distribution, the denoising process can effectively reduce noise without introducing Rician bias.
2Object-affected harmful factors
If strong denoising is applied to reduce noise, then noise levels decrease, but small anatomical features are lost due to over-smoothing
Solution Approach 1:
By changing from magnitude image processing to complex image processing, the invention achieves superior noise reduction performance. The Gaussian noise model in the complex domain allows for more effective denoising that preserves fine anatomical structures better than magnitude-based methods, reducing the need for aggressive filtering that causes over-smoothing.
3Measurement precision
If multiple acquisitions are performed to improve signal-to-noise ratio, then image quality improves, but examination time and cost increase
Solution Approach 1:
The invention replaces the mechanical approach of acquiring multiple images and averaging them with a computational approach using deep learning on complex-valued images. This substitution achieves comparable or superior noise reduction in a single acquisition, eliminating the time penalty of multiple scans.
Data Source
AI summary
The invention relates to a method of MR imaging of an object positioned in the examination volume of an MR system (1). It is an object of the invention to provide a deep learning-based denoising approach that overcomes the Rician bias problem. As a solution, the invention proposes a method comprising the following steps: a) subjecting the object to an imaging sequence comprising RF pulses and switched magnetic field gradients, whereby MR signals are generated, b) acquiring the MR signals, c) reconstructing a complex-valued MR image from the acquired MR signals, d) denoising the MR image using a deep learning algorithm that operates on the real and the imaginary parts of the MR image, and c) computing a magnitude MR image from the denoised complex-valued MR image. According to an aspect of the invention, the deep learning algorithm uses a set of denoising models that are trained using different loss functions. In this way, a trade-off between noise removal and preservation of small image details can be controlled. Moreover, the invention relates to an MR system (1) and to a computer program for an MR system (1).