Related Experiment Video
Updated: Nov 25, 2025
Three-Dimensional Shape Modeling and Analysis of Brain Structures
Published on: November 14, 2019
Joint affine and deformable three-dimensional networks for brain MRI registration
Zhenyu Zhu1, Yiqin Cao1, Chenchen Qin1
1National-Regional Key Technology Engineering Laboratory for Medical Ultrasound, Guangdong Key Laboratory for Biomedical Measurements and Ultrasound Imaging, School of Biomedical Engineering, Health Science Center, Shenzhen University, Shenzhen, China.
This study introduces an end-to-end deep learning network for 3D medical image registration, combining rigid and deformable alignment. The novel approach achieves accurate and robust volumetric registration without pre-alignment, outperforming existing methods.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computational Anatomy
Background:
- Volumetric medical image registration is crucial for clinical applications.
- Traditional methods are slow due to iterative optimization.
- Current deep learning methods often require separate rigid and deformable alignment steps, hindering end-to-end processing.
Purpose of the Study:
- To develop an end-to-end deep learning network for simultaneous affine and deformable 3D medical image registration.
- To overcome the limitations of sequential alignment in existing deep learning registration models.
Main Methods:
- Proposed a novel end-to-end joint affine and deformable network for 3D medical image registration.
- Integrated affine and deformable registration subnetworks with shared parameters.
- Utilized global and local similarity measures as loss functions, augmented by an anatomical similarity loss for weak supervision.
Main Results:
- The network was evaluated on three public brain MRI datasets (Mindboggle101, LPBA40, IXI).
- Demonstrated superior performance compared to state-of-the-art methods.
- Achieved high accuracy based on Dice index (DSC), Hausdorff distance (HD), and average symmetric surface distance (ASSD).
Conclusions:
- The developed network enables accurate and robust volumetric registration.
- Eliminates the need for pre-alignment, facilitating true end-to-end deformable registration.
- Offers a significant advancement in automated medical image analysis.
