A multi-level segmentation-guided diffusion model for streak artifact reduction in routine non-contrast chest CT.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Streak artifacts in non-contrast computed tomography (NCCT) compromise diagnostic accuracy by obscuring anatomical details.
• Existing artifact reduction methods face challenges with scalability, anatomical constraint, and diagnostic applicability due to high annotation costs and limited generative stability.
• There is a need for advanced methods to effectively reduce streak artifacts while preserving crucial anatomical information in NCCT scans.

Purpose of the Study:

• To develop and evaluate a novel guided diffusion method for optimizing streak artifact reduction in chest NCCT scans.
• To leverage multi-level anatomical segmentations to enhance artifact reduction while maintaining structural integrity.
• To improve the clinical utility of NCCT by enhancing medical image quality and diagnostic reliability.

Main Methods:

• A guided diffusion model was developed, integrating artifact-free CT slices with multi-level segmentation maps and anatomical regions of interest (ROIs) during training.
• The model was trained on a large dataset of 96,641 CT slices from four different scanners.
• During inference, artifact-affected samples were processed to generate artifact-free outputs with preserved structural integrity.

Main Results:

• The proposed method demonstrated significant improvements in Signal-to-Noise Ratio (SNR) and Contrast-to-Noise Ratio (CNR) compared to artifact-affected samples (p < 0.05).
• Quantitative assessments showed high consistency between generated outputs and reference artifact-free samples, with no significant difference in lung field SNR and lung-trachea CNR (p > 0.05).
• The method outperformed four novel studies in Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Dice Similarity Coefficient (DSC) (all p < 0.05).

Conclusions:

• The novel guided diffusion method effectively reduces streak artifacts in chest NCCT scans, significantly enhancing image quality.
• Multi-level anatomical segmentation guidance, particularly Level-2, optimally balances anatomical structure preservation and artifact reduction efficiency.
• This technique holds significant potential for improving clinical analysis of chest NCCT by providing clearer, artifact-free images for diagnosis.