Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Iterative Learning for Joint Image Denoising and Motion Artifact Correction of 3D Brain MRI

Lintao Zhang, Mengqi Wu, Lihong Wang, David C. Steffens, Guy G. Potter, Mingxia Liu

TL;DR

Experimental results suggest the effectiveness of JDAC in both tasks of denoising and motion artifact correction, compared with several state-of-the-art methods.

Abstract

Image noise and motion artifacts greatly affect the quality of brain MRI and negatively influence downstream medical image analysis. Previous studies often focus on 2D methods that process each volumetric MR image slice-by-slice, thus losing important 3D anatomical information. Additionally, these studies generally treat image denoising and artifact correction as two standalone tasks, without considering their potential relationship, especially on low-quality images where severe noise and motion artifacts occur simultaneously. To address these issues, we propose a Joint image Denoising and motion Artifact Correction (JDAC) framework via iterative learning to handle noisy MRIs with motion artifacts, consisting of an adaptive denoising model and an anti-artifact model. In the adaptive denoising model, we first design a novel noise level estimation strategy, and then adaptively reduce the noise through a U-Net backbone with feature normalization conditioning on the estimated noise variance. The anti-artifact model employs another U-Net for eliminating motion artifacts, incorporating a novel gradient-based loss function designed to maintain the integrity of brain anatomy during the motion correction process. These two models are iteratively employed for joint image denoising and artifact correction through an iterative learning framework. An early stopping strategy depending on noise level estimation is applied to accelerate the iteration process. The denoising model is trained with 9,544 T1-weighted MRIs with manually added Gaussian noise as supervision. The anti-artifact model is trained on 552 T1-weighted MRIs with motion artifacts and paired motion-free images. Experimental results on a public dataset and a clinical study suggest the effectiveness of JDAC in both tasks of denoising and motion artifact correction, compared with several state-of-the-art methods.

Iterative Learning for Joint Image Denoising and Motion Artifact Correction of 3D Brain MRI

TL;DR

Experimental results suggest the effectiveness of JDAC in both tasks of denoising and motion artifact correction, compared with several state-of-the-art methods.

Abstract

Image noise and motion artifacts greatly affect the quality of brain MRI and negatively influence downstream medical image analysis. Previous studies often focus on 2D methods that process each volumetric MR image slice-by-slice, thus losing important 3D anatomical information. Additionally, these studies generally treat image denoising and artifact correction as two standalone tasks, without considering their potential relationship, especially on low-quality images where severe noise and motion artifacts occur simultaneously. To address these issues, we propose a Joint image Denoising and motion Artifact Correction (JDAC) framework via iterative learning to handle noisy MRIs with motion artifacts, consisting of an adaptive denoising model and an anti-artifact model. In the adaptive denoising model, we first design a novel noise level estimation strategy, and then adaptively reduce the noise through a U-Net backbone with feature normalization conditioning on the estimated noise variance. The anti-artifact model employs another U-Net for eliminating motion artifacts, incorporating a novel gradient-based loss function designed to maintain the integrity of brain anatomy during the motion correction process. These two models are iteratively employed for joint image denoising and artifact correction through an iterative learning framework. An early stopping strategy depending on noise level estimation is applied to accelerate the iteration process. The denoising model is trained with 9,544 T1-weighted MRIs with manually added Gaussian noise as supervision. The anti-artifact model is trained on 552 T1-weighted MRIs with motion artifacts and paired motion-free images. Experimental results on a public dataset and a clinical study suggest the effectiveness of JDAC in both tasks of denoising and motion artifact correction, compared with several state-of-the-art methods.
Paper Structure (27 sections, 17 equations, 8 figures, 6 tables, 1 algorithm)

This paper contains 27 sections, 17 equations, 8 figures, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. MR Image Denoising
  4. Motion Artifact Correction of MRI
  5. Methodology
  6. Proposed Method
  7. Proposed Iterative Learning Framework
  8. Iterative Learning Strategy
  9. Noise Estimation Strategy of Structural MRI
  10. Adaptive Denoising Model
  11. Anti-artifact Model
  12. Experiments
  13. Experimental Setup
  14. Competing Methods
  15. Joint Denoising and Motion Correction on MR-ART
  16. ...and 12 more sections

Figures (8)

  • Figure 1: Illustration of the proposed iterative learning framework ( i.e., panel (a)) of joint image denoising and motion artifact correction (JDAC) for structural MRI data processing. The JDAC consists of an adaptive denoising model ( i.e., panel (b)) and an anti-artifact model ( i.e., panel (c)) that iteratively reduces the MRI image noise and motion artifacts. The denoising model can adaptively denoise the noisy MRI based on estimated noise levels. The anti-artifact model is trained with motion-free MRIs as ground truth and constrained by a new gradient-based loss function for brain structure preserving.
  • Figure 2: Illustration of an original motion-affected MRI from MR-ART narai2022movement with different manually added Gaussian noise levels (top) and their corresponding gradient maps (bottom). The contrast of gradient maps is normalized for better visualization. Std: standard deviation.
  • Figure 3: Average standard deviation (std) values of perturbed MR images ( i.e., MRI with manually added Gaussian noise), their corresponding gradient maps, and gradient maps of the added Gaussian noise. These averaged values are calculated based on 40 MRIs randomly selected from (a) ADNI jack2008alzheimer, (b) MR-ART narai2022movement, and (c) NBOLD steffens2017negative.
  • Figure 4: Qualitative comparison of the JDAC against the six competing methods ( i.e., DRN-DCMB, SUNet, BM4D, UNet3D, nnUNet, and FONDUE) on one subject (ID: 862915) in the MR-ART narai2022movement dataset with manually added Gaussian noise (std: 0.10). The panel (a) shows the denoising and anti-artifact results of MRI with minor head motion, and the panel (b) shows results of the same subject with excessive head motion.
  • Figure 5: Real motion cases of three subjects in the NBOLD study steffens2017negative. The first column images are the motion-affected real clinical MRIs, while the other columns are the denoised and motion-corrected images via the competing methods and the proposed JDAC. The circles show areas with significantly different denoising results, and the rectangles highlight areas with different artifact correction results.
  • ...and 3 more figures