Related Experiment Video
Updated: Jul 19, 2025
Multi-modal Pulmonary Imaging: Using Complementary Information from CT and Hyperpolarized 129Xe MRI to Evaluate Lung Structure-Function
Published on: April 12, 2024
MD-SGT: Multi-dilation spherical graph transformer for unsupervised medical image registration
Kun Tang1, Lihui Wang1, Xingyu Huang1
1Engineering Research Center of Text Computing & Cognitive Intelligence, Ministry of Education, Key Laboratory of Intelligent Medical Image Analysis and Precise Diagnosis of Guizhou Province, State Key Laboratory of Public Big Data, College of Computer Science and Technology, Guizhou University, Guiyang, China.
This study introduces a novel multi-dilation spherical graph transformer (MD-SGT) for deformable medical image registration. The MD-SGT improves registration accuracy by enhancing long-range spatial dependence and attention span.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Deformable medical image registration is crucial for clinical applications.
- Current convolutional neural network and transformer methods have limitations in capturing long-range spatial dependencies and uniform attention.
Purpose of the Study:
- To address the limitations of existing methods in deformable medical image registration.
- To propose a novel model that enhances long-range spatial dependence and attention span for improved registration performance.
Main Methods:
- Introduced a multi-dilation spherical graph transformer (MD-SGT) model.
- The encoder integrates convolutional and graph transformer blocks for multi-scale feature extraction.
- Voxel features are generated by aggregating neighbor information from spherical regions with varying dilation rates.
Main Results:
- The MD-SGT model demonstrated improved performance over state-of-the-art methods on two datasets.
- Key metrics showed significant improvements: Dice score (≥0.5%), ASD (≥2.2%), and HD95 (≥1.1%).
- The study validates the benefits of combining long-range uniform attention span and inductive bias.
Conclusions:
- The proposed MD-SGT effectively captures multi-scale features and long-range dependencies for medical image registration.
- Integrating inductive bias and uniform attention span enhances feature representativeness and registration accuracy.
- The findings suggest a promising direction for advancing deformable medical image registration techniques.
