A Guided Refinement Network Model With Joint Denoising and Segmentation for Low-Dose Coronary CTA Subtle Structure...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Disease

Background:

• Coronary CT angiography (CCTA) is vital for assessing coronary arteries but involves ionizing radiation risks, including cancer.
• Low-dose CCTA reduces radiation but compromises image quality, hindering diagnosis due to noise and subtle structure degradation.
• Existing deep learning methods struggle to restore fine coronary branches without location priors, leading to boundary ambiguity.

Purpose of the Study:

• To develop a novel deep learning model for high-quality image restoration from low-dose CCTA.
• To address the limitations of existing methods in noise suppression and subtle structure preservation.
• To improve diagnostic accuracy in low-dose CCTA by enhancing image quality and coronary visualization.

Main Methods:

• Proposes a Guided Refinement Network (GRN) model utilizing joint learning for denoising and coronary segmentation.
• Integrates coronary segmentation to provide location priors for denoising, guiding the preservation of subtle structures.
• Employs mutual guidance between denoising and segmentation for collaborative optimization and improved results.

Main Results:

• GRN effectively suppresses noise and enhances subtle coronary structures in low-dose CCTA images.
• The model outperforms existing methods in quantitative and qualitative assessments of image quality and structure restoration.
• GRN generates reliable coronary segmentation masks, valuable for diagnostic assistance.

Conclusions:

• GRN successfully restores high-quality images from low-dose CCTA through joint denoising and segmentation.
• The proposed method significantly improves noise suppression and subtle structure restoration compared to current techniques.
• GRN provides valuable diagnostic support for radiologists by offering enhanced coronary visualization and segmentation masks.