Related Experiment Video
Updated: Aug 4, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
Unsupervised Cross-Modality Adaptation via Dual Structural-Oriented Guidance for 3D Medical Image Segmentation
IEEE Transactions on Medical Imaging
|April 6, 2023
Summary
This study introduces DAG-Net, a novel unsupervised domain adaptation method for medical image segmentation. DAG-Net improves deep learning model performance on diverse, unseen data by focusing on structural features and geometric continuity.
Area of Science:
- Artificial Intelligence
- Medical Imaging
- Computer Vision
Background:
- Deep convolutional neural networks (CNNs) excel in medical image segmentation but struggle with data heterogeneity.
- Unsupervised domain adaptation (UDA) is crucial for adapting models to unseen data without labels.
Purpose of the Study:
- To develop a novel UDA method, DAG-Net, for robust medical image segmentation across different data characteristics.
- To enhance the adaptability of segmentation models to unlabeled target domains.
Main Methods:
- Introduced Dual Adaptation-Guiding Network (DAG-Net) with two core modules: Fourier-based contrastive style augmentation (FCSA) and residual space alignment (RSA).
- FCSA learns modality-insensitive, structure-relevant features.
- RSA enhances prediction geometric continuity using 3D inter-slice correlation priors.
Main Results:
- DAG-Net demonstrated superior performance in cross-modality adaptation between MRI and CT for cardiac and abdominal organ segmentation.
- Outperformed state-of-the-art UDA methods on unlabeled target 3D medical images.
Conclusions:
- DAG-Net effectively addresses domain shift challenges in 3D medical image segmentation.
- The proposed method offers a robust solution for adapting segmentation models to heterogeneous, unlabeled medical imaging data.
