Related Experiment Video
Updated: Jul 11, 2025
10:39
A Label-Free Segmentation Approach for Intravital Imaging of Mammary Tumor Microenvironment
Published on: May 24, 2022
2.4K
C-DARL: Contrastive diffusion adversarial representation learning for label-free blood vessel segmentation
Boah Kim1, Yujin Oh2, Bradford J Wood1
1Radiology and Imaging Sciences, National Institutes of Health Clinical Center, Bethesda, MD, USA.
Medical Image Analysis
|November 17, 2023
Summary
This study introduces C-DARL, a novel self-supervised method for blood vessel segmentation in medical images. The model effectively generates realistic vessel representations, improving diagnostic accuracy for vascular diseases.
Area of Science:
- Medical Imaging
- Computer Vision
- Artificial Intelligence
Background:
- Manual blood vessel segmentation is labor-intensive and complex.
- Accurate segmentation is crucial for diagnosing vascular diseases and planning interventions.
Purpose of the Study:
- To develop a self-supervised method for automated blood vessel segmentation.
- To improve the realism and accuracy of vessel representations in medical images.
Main Methods:
- Introduced the contrastive diffusion adversarial representation learning (C-DARL) model.
- Utilized diffusion and generation modules for synthetic vessel data.
- Employed contrastive learning with a mask-based contrastive loss.
Main Results:
- C-DARL demonstrated improved performance over baseline methods.
- The model showed robustness to noise in various vessel datasets.
- Achieved effective vessel segmentation in coronary angiograms, abdominal digital subtraction angiograms, and retinal imaging.
Conclusions:
- C-DARL offers an effective solution for self-supervised vessel segmentation.
- The method enhances the learning of realistic vessel representations.
- This approach holds promise for improving vascular disease diagnosis and interventional planning.
