Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DDO-IN: Dual Domains Optimization for Implicit Neural Network to Eliminate Motion Artifact in Magnetic Resonance Imaging

Zhongyu Mai, Zewei Zhan, Hanyu Guo, Yulang Huang, Weifeng Su

TL;DR

DDO-IN tackles MRI motion artifacts by fusing frequency-domain (k-space) and pixel-domain information through implicit neural representations. The approach employs a Deartifact INR and a Movement INR, guided by a kMR-Net mask, and uses a dynamic loss weight that shifts emphasis from global to local restoration, enabling robust artifact removal and texture preservation. Dual-domain fusion combines low-frequency $f_{cl}$ with high-frequency content via a mask $\mathscr{M}$ to produce $f_{ic}$, from which $i_o = \mathscr{F}^{-1}(f_{ic})$ is obtained; this is optimized with a scheduled loss that balances frequency and pixel domains. On NYU fastMRI brain data, DDO-IN achieves state-of-the-art performance across VIF, HaarPSI, PSNR, and SSIM for both mild and severe motion, demonstrating strong practical potential for improved diagnostic quality and reduced scan repeats.

Abstract

Magnetic resonance imaging (MRI) motion artifacts can seriously affect clinical diagnostics, making it challenging to interpret images accurately. Existing methods for eliminating motion artifacts struggle to retain fine structural details and simultaneously lack the necessary vividness and sharpness. In this study, we present a novel dual-domain optimization (DDO) approach that integrates information from the pixel and frequency domains guiding the recovery of clean magnetic resonance images through implicit neural representations(INRs). Specifically, our approach leverages the low-frequency components in the k-space as a reference to capture accurate tissue textures, while high-frequency and pixel information contribute to recover details. Furthermore, we design complementary masks and dynamic loss weighting transitioning from global to local attention that effectively suppress artifacts while retaining useful details for reconstruction. Experimental results on the NYU fastMRI dataset demonstrate that our method outperforms existing approaches in multiple evaluation metrics. Our code is available at https://anonymous.4open.science/r/DDO-IN-A73B.

DDO-IN: Dual Domains Optimization for Implicit Neural Network to Eliminate Motion Artifact in Magnetic Resonance Imaging

TL;DR

DDO-IN tackles MRI motion artifacts by fusing frequency-domain (k-space) and pixel-domain information through implicit neural representations. The approach employs a Deartifact INR and a Movement INR, guided by a kMR-Net mask, and uses a dynamic loss weight that shifts emphasis from global to local restoration, enabling robust artifact removal and texture preservation. Dual-domain fusion combines low-frequency with high-frequency content via a mask to produce , from which is obtained; this is optimized with a scheduled loss that balances frequency and pixel domains. On NYU fastMRI brain data, DDO-IN achieves state-of-the-art performance across VIF, HaarPSI, PSNR, and SSIM for both mild and severe motion, demonstrating strong practical potential for improved diagnostic quality and reduced scan repeats.

Abstract

Magnetic resonance imaging (MRI) motion artifacts can seriously affect clinical diagnostics, making it challenging to interpret images accurately. Existing methods for eliminating motion artifacts struggle to retain fine structural details and simultaneously lack the necessary vividness and sharpness. In this study, we present a novel dual-domain optimization (DDO) approach that integrates information from the pixel and frequency domains guiding the recovery of clean magnetic resonance images through implicit neural representations(INRs). Specifically, our approach leverages the low-frequency components in the k-space as a reference to capture accurate tissue textures, while high-frequency and pixel information contribute to recover details. Furthermore, we design complementary masks and dynamic loss weighting transitioning from global to local attention that effectively suppress artifacts while retaining useful details for reconstruction. Experimental results on the NYU fastMRI dataset demonstrate that our method outperforms existing approaches in multiple evaluation metrics. Our code is available at https://anonymous.4open.science/r/DDO-IN-A73B.

Paper Structure

This paper contains 10 sections, 10 equations, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methodology
  3. Implicit Neural Network
  4. Dual Domain Optimization Pipeline
  5. Dynamic Loss Function
  6. Experimental result
  7. Datasets & Setup
  8. Results
  9. Discussion & Outlook
  10. Conclusion

Figures (2)

  • Figure 1: The framework of DDO-IN. The red area corresponds to the INN module, responsible for generating the deartifacted image and motion transformation map. These two outputs are subsequently fused via the kMR-Net. The green area illustrates frequency domain processing, which combines low-frequency and high-frequency information. The yellow area indicates pixel domain processing, capturing pixel-level details. After merging the results from both domains, the clean image is generated by Deartifact INR through iterative optimization of the loss function.
  • Figure 2: The illustrated data demonstrates the median results of motion-corrected images from our DDO-IN pipeline, alongside motion-corrupted, ground truth, and comparison methods. The first and third rows present the light and heavy correction outcomes, respectively. The second and fourth rows illustrate the residual error images.