Pre- to post-contrast medical image synthesis with outline-guide accelerate diffusion model

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Contrast agents enhance medical image visibility but can cause adverse reactions.
• Generative models aim to synthesize post-contrast images from pre-contrast scans to avoid contrast agent risks.
• Diffusion models show promise for image synthesis but face challenges with anatomical deformation and slow processing.

Purpose of the Study:

• To develop an improved diffusion model for synthesizing post-contrast medical images from pre-contrast images.
• To address limitations of existing diffusion models, including anatomical structure deformation and slow synthesis speeds.
• To enhance the accuracy and efficiency of generating contrast-enhanced medical images without contrast agents.

Main Methods:

• Proposed a novel outline-guide accelerated diffusion model for medical image synthesis.
• Utilized outline information from pre-contrast images to ensure anatomical structure consistency.
• Incorporated a multi-frequency enhanced attention module for improved feature distinction and a non-uniform sampling strategy to accelerate synthesis.

Main Results:

• The proposed model maintains clearer detail texture compared to existing methods.
• Achieved high-quality post-contrast medical image synthesis with significantly reduced training time.
• Demonstrated average improvements: 10.2% SSIM, 43.7% PSNR, 9.5% MSIM, and a 58.4% NRMSE reduction.

Conclusions:

• The outline-guide accelerated diffusion model effectively synthesizes post-contrast medical images with improved anatomical consistency and detail.
• The model offers a faster and more accurate alternative to traditional contrast-enhanced imaging, reducing risks associated with contrast agents.
• This approach holds significant potential for enhancing medical diagnostics and patient safety.