Related Experiment Video
Updated: Jul 5, 2025
07:13
Author Spotlight: An Efficient and Robust Software for Automated Fusion of Multiple Preclinical Imaging Modalities
Published on: October 27, 2023
1.2K
MAIRNet: weakly supervised anatomy-aware multimodal articulated image registration network
Xiaoru Gao1, Woquan Zhong2, Runze Wang1
1Institute of Medical Robotics, Shanghai Jiao Tong University, Dongchuan Road, Shanghai, 200240, China.
International Journal of Computer Assisted Radiology and Surgery
|January 18, 2024
Summary
This study introduces MAIRNet, a new network for multimodal articulated image registration (MAIR). MAIRNet effectively handles both rigid and deformable structures, outperforming existing methods on challenging datasets.
Area of Science:
- Medical Imaging
- Computer Vision
- Biomedical Engineering
Background:
- Multimodal articulated image registration (MAIR) is complex due to the need to preserve rigidity in bones while allowing soft tissue deformation.
- Current deep learning methods often treat MAIR as purely deformable registration, neglecting articulated structures and yielding suboptimal outcomes.
Purpose of the Study:
- To develop a novel weakly supervised, anatomy-aware network (MAIRNet) for multimodal articulated image registration.
- To address the limitations of existing methods by explicitly handling articulated structures in MAIR.
Main Methods:
- MAIRNet features a dual-branch architecture: a non-learnable polyrigid branch for initial velocity field estimation and a learnable deformable branch for incremental updates.
- These branches collaborate to generate a comprehensive velocity field, which is then integrated to produce the final displacement field for registration.
Main Results:
- Experiments on hip, lumbar, and thoracic spine datasets demonstrated MAIRNet's effectiveness.
- Achieved high Dice scores, e.g., 90.8% for pelvis, 86.1% for L4 vertebrae, and 85.7% for T10 vertebrae.
- Outperformed state-of-the-art methods in multimodal articulated image registration tasks.
Conclusions:
- A novel approach, MAIRNet, was developed for multimodal articulated image registration.
- Extensive validation on diverse datasets confirmed the method's efficacy.
- MAIRNet surpasses current state-of-the-art techniques in MAIR.
