CT-SDM: A Sampling Diffusion Model for Sparse-View CT Reconstruction Across Various Sampling Rates.

Area of Science:

• Medical Imaging
• Computational Imaging
• Radiology

Background:

• Sparse-view computed tomography (SVCT) reduces radiation dose but introduces artifacts due to limited projection views.
• Traditional reconstruction methods struggle with artifacts in SVCT.
• Deep learning shows promise for SVCT artifact removal but lacks generalization across sampling rates.

Purpose of the Study:

• To develop an adaptive reconstruction method for high-performance SVCT across diverse sampling rates.
• To enhance the usability and flexibility of deep learning models in clinical SVCT settings.

Main Methods:

• Proposed a novel sampling diffusion model for SVCT (CT-SDM).
• Designed a unique imaging degradation operator to simulate the sinogram projection process.
• Enabled gradual addition of projection views from undersampled measurements to full-view sinograms.

Main Results:

• The CT-SDM generalizes to various sampling rates using a single trained model.
• Demonstrated effective and robust high-quality image reconstruction from sparse-view CT scans.
• Achieved superior performance compared to existing methods across different sampling rates.

Conclusions:

• The proposed adaptive reconstruction method significantly improves SVCT image quality.
• CT-SDM offers a flexible and effective solution for clinical applications requiring variable sampling rates.
• This approach overcomes the generalization limitations of current deep learning models for SVCT.