DiffusionCT: Latent Diffusion Model for CT Image Standardization

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Computed tomography (CT) is crucial for lung cancer management, but variations in scanner protocols create inconsistent image textures.
• These texture variations pose a significant challenge for reliable quantitative analysis in downstream studies.
• Current CT image harmonization methods, often GAN-based, show limited effectiveness.

Purpose of the Study:

• To develop a novel diffusion-based model, DiffusionCT, for harmonizing CT images acquired from different vendors and protocols.
• To address the limitations of existing supervised and semi-supervised learning approaches for CT image standardization.
• To improve the consistency and reliability of texture features extracted from CT scans for clinical applications.

Main Methods:

• DiffusionCT employs a U-Net-based encoder-decoder architecture operating in the latent space.
• A diffusion model is integrated into the bottleneck of the encoder-decoder.
• The model undergoes a two-phase training process: first training the encoder-decoder, then training the latent diffusion model.

Main Results:

• DiffusionCT successfully maps latent non-standard image distributions to a standard one.
• The model synthesizes standardized CT images from transformed latent representations.
• Experimental results show significant performance improvements in CT image standardization tasks.

Conclusions:

• DiffusionCT offers a promising new approach for harmonizing CT images, overcoming limitations of previous methods.
• The diffusion-based latent space harmonization enhances the reliability of quantitative analysis in medical imaging.
• This advancement has the potential to improve the accuracy of lung cancer screening, diagnosis, and treatment monitoring.