GANs-guided Conditional Diffusion Model for Synthesizing Contrast-enhanced Computed Tomography Images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Contrast-enhanced computed tomography (CE-CT) is vital for diagnosing focal liver lesions but poses burdens.
• Generative Adversarial Networks (GANs) and Diffusion Models (DMs) show promise for synthesizing CE-CT from non-contrast CT (NC-CT) images.
• Existing methods like GANs suffer from coverage and mode collapse, while DMs yield lower local quality, critical for lesion detection.

Purpose of the Study:

• To develop an advanced model for synthesizing high-quality CE-CT images from NC-CT scans.
• To address the limitations of existing GANs and DMs in medical image synthesis.
• To improve the diagnostic accuracy of focal liver lesions by enhancing CE-CT image quality.

Main Methods:

• Proposed a novel GANs-guided conditional diffusion model (GANs-CDM).
• Utilized GANs to generate preliminary CE-CT images as conditional input.
• Employed a conditional diffusion model (CDM) for refining the synthesized CE-CT images.
• Evaluated performance on arterial and portal venous phase synthesis tasks.

Main Results:

• The GANs-CDM significantly improved both local and global quality of synthesized CE-CT images.
• Demonstrated superior performance compared to existing methods in qualitative and quantitative evaluations.
• Successfully generated CE-CT images with enhanced detail in lesion areas.

Conclusions:

• The proposed GANs-CDM effectively overcomes the limitations of previous GANs and DMs for medical image synthesis.
• This approach offers a promising alternative to traditional CE-CT scans, reducing patient burden.
• Enhanced synthetic CE-CT image quality supports more accurate diagnosis of focal liver lesions.