Related Experiment Video
Updated: Dec 22, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
697
Unsupervised Domain Adaptation via Disentangled Representations: Application to Cross-Modality Liver Segmentation.
Junlin Yang1, Nicha C Dvornek2, Fan Zhang3
1Department of Biomedical Engineering, Yale University, New Haven, CT, USA.
Medical Image Computing and Computer-Assisted Intervention : MICCAI ... International Conference on Medical Image Computing and Computer-Assisted Intervention
|May 8, 2020
Summary
This study introduces a novel deep learning approach for cross-modality domain adaptation, improving medical image segmentation between CT and MRI. The method achieves superior performance by disentangling content and style, outperforming existing techniques.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Deep learning models struggle with domain shifts, performing poorly on data from different sources like CT and MRI.
- Unsupervised domain adaptation (UDA) methods aim to bridge this gap using labeled source data and unlabeled target data.
Purpose of the Study:
- To develop a cross-modality domain adaptation technique for medical imaging, specifically between CT and MRI.
- To improve the performance of deep learning models on tasks like liver segmentation across different imaging modalities.
Main Methods:
- A novel method using disentangled representations to achieve cross-modality domain adaptation.
- Embedding images into a shared domain-invariant content space and a domain-specific style space.
- Recovering a many-to-many mapping between domains to capture complex cross-domain relations, preserving semantic feature-level information.
Main Results:
- Achieved a Dice Similarity Coefficient (DSC) of 0.81 for cross-modality liver segmentation (CT to MRI), outperforming CycleGAN (0.72).
- Demonstrated good generalization to joint-domain learning, improving segmentation on individual modalities.
- Showcased potential for diverse image generation and effectiveness on multi-phasic MRI (DSC 0.74) compared to CycleGAN (DSC 0.52).
Conclusions:
- The proposed disentangled representation method effectively addresses cross-modality domain adaptation challenges in medical imaging.
- This approach offers a robust solution for improving segmentation performance and enabling diverse image generation across different modalities.
