Related Experiment Video
Updated: Sep 18, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
Generative deep-learning-model based contrast enhancement for digital subtraction angiography using a
Takeshi Takata1, Kentaro Yamada2, Masayoshi Yamamoto3
1Advanced Comprehensive Research Organization, Teikyo University, Tokyo, 173-0003, Japan.
Computers in Biology and Medicine
|June 21, 2025
Summary
Generative deep learning enhances digital subtraction angiography (DSA) contrast, improving vascular visualization. This AI technique offers potential for clearer imaging, especially for patients needing reduced contrast agents.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Radiology
- Interventional Radiology
Background:
- Digital subtraction angiography (DSA) is crucial for visualizing blood vessels.
- Reduced contrast in DSA can obscure vascular structures, a significant issue for patients with chronic kidney disease (CKD) requiring minimized contrast agents to prevent contrast-induced nephropathy (CIN).
Purpose of the Study:
- To explore a generative deep learning model for contrast enhancement in DSA.
- To improve the clarity of vascular structures in DSA images, particularly in low-contrast scenarios.
Main Methods:
- Developed a text-conditioned image-to-image model using Stable Diffusion with ControlNet and Low-Rank Adaptation.
- Trained and tested the model on 1207 DSA series, using data augmentation for low-contrast images.
- Evaluated performance using metrics including RMS contrast, Michelson contrast, SNR, and entropy.
Main Results:
- Significant improvements observed in RMS contrast (7.91 to 17.7), Michelson contrast (0.875 to 0.992), and entropy (3.60 to 5.60), indicating enhanced detail.
- Signal-to-noise ratio (SNR) decreased from 21.3 to 8.50, suggesting an increase in image noise.
Conclusions:
- Deep learning-based contrast enhancement is feasible for DSA.
- Generative deep learning models show promise for improving angiographic imaging quality.
- Further research is needed for artifact suppression and clinical validation for practical application.
