Related Experiment Video
Updated: Nov 15, 2025
06:56
Human Fetal Blood Flow Quantification with Magnetic Resonance Imaging and Motion Compensation
Published on: January 7, 2021
2.6K
Anatomy-Guided Convolutional Neural Network for Motion Correction in Fetal Brain MRI
Yuchen Pei1,2, Lisheng Wang1, Fenqiang Zhao2
1Institute of Image Processing and Pattern Recognition, Department of Automation, Shanghai Jiao Tong University, Shanghai, China.
Machine Learning in Medical Imaging. MLMI (Workshop)
|March 1, 2021
Summary
This study introduces a novel multi-task learning framework for fetal Magnetic Resonance Imaging (MRI). The method improves 3D motion correction by jointly learning transformation parameters and tissue segmentation, enhancing fetal brain MRI reconstruction.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuroscience
Background:
- Fetal MRI is crucial for diagnosing developmental abnormalities but is significantly hindered by fetal motion and maternal breathing.
- Existing motion correction methods for 2D slices are operator-dependent and time-consuming, impacting 3D reconstruction accuracy.
- Convolutional neural network (CNN) approaches show promise but often neglect valuable brain structural information.
Purpose of the Study:
- To develop an advanced multi-task learning framework for fetal MRI motion correction.
- To integrate brain anatomical information into the motion correction process for improved 3D reconstruction.
- To enhance the accuracy of predicting 3D motion parameters and simultaneously achieve precise tissue segmentation.
Main Methods:
- A novel multi-task learning framework was proposed, combining transformation parameter prediction and tissue segmentation.
- A two-stage approach was employed: a coarse stage learning shared features, followed by a refinement stage using signed distance maps.
- Signed distance maps, derived from segmentation, were incorporated to guide both motion parameter regression and segmentation tasks.
Main Results:
- The proposed method demonstrated superior performance in reducing motion prediction error compared to state-of-the-art techniques.
- Satisfactory tissue segmentation results were achieved concurrently with motion correction improvements.
- The integration of anatomical information via signed distance maps significantly enhanced both regression and segmentation tasks.
Conclusions:
- The developed multi-task learning framework effectively addresses the challenges of fetal MRI motion correction.
- Jointly learning motion parameters and tissue segmentation, guided by anatomical information, leads to more accurate 3D fetal brain MRI reconstruction.
- This approach offers a more robust and efficient solution for clinical applications of fetal MRI.
