Related Experiment Video
Updated: Jan 18, 2026
07:13
Author Spotlight: An Efficient and Robust Software for Automated Fusion of Multiple Preclinical Imaging Modalities
Published on: October 27, 2023
1.7K
UTMorph: A hybrid CNN-transformer network for weakly-supervised multimodal image registration in biopsy puncture
Xudong Guo1, Peiyu Chen1, Haifeng Wang2
1School of Health Science and Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China.
Medical Image Analysis
|January 15, 2026
Summary
UTMorph, a novel framework using CNN and Transformer networks, accurately registers preoperative MRI and intraoperative ultrasound images. This improves robotic-guided procedures by precisely handling soft tissue deformation.
Area of Science:
- Medical image analysis
- Artificial intelligence in medicine
- Surgical robotics
Background:
- Accurate registration of preoperative magnetic resonance imaging (MRI) and intraoperative ultrasound (US) is crucial for precise robotic-guided biopsy and ablation.
- Existing methods struggle with speed, accuracy, and soft tissue deformation during procedures.
Purpose of the Study:
- To develop an efficient and accurate deformable multimodal image registration framework for MRI-US fusion.
- To address the challenges of soft tissue deformation and computational complexity in medical image registration.
Main Methods:
- Proposed UTMorph, a hybrid framework combining U-Net, convolutional neural networks (CNN), and Transformer networks.
- Introduced a novel attention mechanism focusing on structured image features for precise deformation estimation.
- Developed a hybrid edge loss function to enhance shape and boundary information for improved registration accuracy.
Main Results:
- UTMorph demonstrated superior performance across multiple metrics compared to six common registration methods.
- Achieved a Dice Similarity Coefficient of 0.890, 95th percentile Hausdorff distance of 2.679 mm, and mean surface distance of 0.284 mm.
- Ensured accurate registration with minimal memory usage, even with significant modal differences, validated on 704 patient datasets.
Conclusions:
- The UTMorph model with its hybrid edge loss function is effective for MR-US deformable medical image registration.
- This framework enhances precision in robotic-assisted interventions by improving image registration accuracy and speed.
- The developed model offers a robust solution for multimodal medical image registration challenges.
