Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Literature Review on Fetus Brain Motion Correction in MRI

Haoran Zhang, Yun Wang

TL;DR

Fetal brain MRI suffers from motion artifacts that degrade image quality and hinder clinical analysis. The paper surveys two major directions—SVR-based slice-to-volume reconstruction and deep learning methods (CNNs, LSTMs, Transformers, GANs, and diffusion models)—to correct motion and reconstruct high-quality volumes. It highlights open-source tools like NiftyMIC and discusses the promise of transformers and diffusion models for handling large, non-rigid motion, while noting concerns about transparency and interpretability. The review points to near-term opportunities in atlas-informed, physics-aware, and real-time motion correction to improve prenatal brain imaging.

Abstract

This paper provides a comprehensive review of the latest advancements in fetal motion correction in MRI. We delve into various contemporary methodologies and technological advancements aimed at overcoming these challenges. It includes traditional 3D fetal MRI correction methods like Slice to Volume Registration (SVR), deep learning-based techniques such as Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) Networks, Transformers, Generative Adversarial Networks (GANs) and most recent advancements of Diffusion Models. The insights derived from this literature review reflect a thorough understanding of both the technical intricacies and practical implications of fetal motion in MRI studies, offering a reasoned perspective on potential solutions and future improvements in this field.

A Literature Review on Fetus Brain Motion Correction in MRI

TL;DR

Fetal brain MRI suffers from motion artifacts that degrade image quality and hinder clinical analysis. The paper surveys two major directions—SVR-based slice-to-volume reconstruction and deep learning methods (CNNs, LSTMs, Transformers, GANs, and diffusion models)—to correct motion and reconstruct high-quality volumes. It highlights open-source tools like NiftyMIC and discusses the promise of transformers and diffusion models for handling large, non-rigid motion, while noting concerns about transparency and interpretability. The review points to near-term opportunities in atlas-informed, physics-aware, and real-time motion correction to improve prenatal brain imaging.

Abstract

This paper provides a comprehensive review of the latest advancements in fetal motion correction in MRI. We delve into various contemporary methodologies and technological advancements aimed at overcoming these challenges. It includes traditional 3D fetal MRI correction methods like Slice to Volume Registration (SVR), deep learning-based techniques such as Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) Networks, Transformers, Generative Adversarial Networks (GANs) and most recent advancements of Diffusion Models. The insights derived from this literature review reflect a thorough understanding of both the technical intricacies and practical implications of fetal motion in MRI studies, offering a reasoned perspective on potential solutions and future improvements in this field.
Paper Structure (13 sections, 5 equations, 2 figures)

This paper contains 13 sections, 5 equations, 2 figures.

Table of Contents

  1. Introduction
  2. Literature Search
  3. Slice-to-Volume Registration
  4. Deep Learning Methods
  5. Convolutional Neural Networks
  6. Long Short-Term Memory Networks
  7. Transformers
  8. Generative Adversarial Networks
  9. Diffusion Model
  10. Variational Perspective
  11. Score Perspective
  12. Application in Motion Correction
  13. Conclusions

Figures (2)

  • Figure 1: 3D SVR reconstruction pipeline for fetal brain MRI. This figure is based on the MRI dataset from St.Thomas’ Hospital, London uus2023retrospective, see the original paper at https://pubmed.ncbi.nlm.nih.gov/35834425.
  • Figure 2: A flowchart depicting CNN-based motion correction techniques from the work in chang2023deep. Initially, motion-corrupted images are input into a network, which then outputs motion-free images. For training the deep learning model, numerous pairs of motion-corrupted and motion-free images are processed through the network. Subsequently, in the correction phase, a motion-corrected image is produced by inputting a motion-corrupted image. This process ultimately results in the generation of a motion-free image, see the original paper at https://www.sciencedirect.com/science/article/pii/S2950162823000012.