Accelerating direct material decomposition via diffusion probabilistic model for Sparse-view spectral computed...

Area of Science:

• Medical Imaging
• Computational Imaging
• Artificial Intelligence in Medicine

Background:

• Accurate material decomposition is crucial for Spectral CT applications.
• Conventional methods struggle with sparse-view artifacts, slow convergence, noise, and ill-posedness, especially in geometrically inconsistent imaging.

Purpose of the Study:

• To develop an unsupervised deep learning framework for direct material decomposition in sparse-view Spectral CT.
• To overcome limitations of conventional model-based methods using virtual monochromatic images (VMIs) and probabilistic diffusion models.

Main Methods:

• Proposed an unsupervised deep learning framework optimizing VMIs via a probabilistic diffusion model for material decomposition.
• Integrated VMIs to enhance differentiation of polychromatic projections and address iterative reconstruction convergence issues.
• Employed dual constraints: data fidelity for measurement consistency and probabilistic regularization for anatomical plausibility.

Main Results:

• Achieved a 10 dB improvement in peak-signal-to-noise ratio (PSNR) and a 4.31% increase in structural similarity (SSIM) for soft-tissue reconstructions compared to existing methods with 90 projections.
• Demonstrated robustness and maintained reconstruction fidelity under sparse sampling and geometric inconsistency.
• Validated effectiveness on preclinical data.

Conclusions:

• The proposed deep learning framework significantly improves material decomposition in sparse-view Spectral CT.
• The method offers enhanced accuracy and robustness, overcoming key challenges in current techniques.
• This approach holds promise for advancing Spectral CT applications requiring high-fidelity material decomposition.