One-Step Material Decomposition Using Spectral Diffusion Posterior Sampling in Sparse-View Dual-Layer CT

Area of Science:

• Medical Imaging
• Computational Imaging
• Materials Science

Background:

• Spectral X-ray imaging offers material-specific information crucial for interventional procedures.
• Cone beam CT (CBCT) systems face speed and scan range limitations, making sparse-view imaging desirable for faster, lower-dose scans.
• Sparse data in CBCT present significant challenges for precise material quantification.

Purpose of the Study:

• To evaluate the performance of the Spectral Diffusion Posterior Sampling (Spectral DPS) framework for material decomposition using sparse-view data.
• To assess Spectral DPS's accuracy and robustness on a physical dual-layer CBCT system under varying degrees of data sparsity.
• To investigate the potential of Spectral DPS in improving interventional CBCT imaging.

Main Methods:

• The Spectral DPS framework, integrating a learned diffusion prior with a physics-based forward model, was applied for one-step material decomposition.
• An anthropomorphic head phantom was scanned on a dual-layer CBCT system with projection views reduced from 720 to 45.
• Material decomposition was performed using Spectral DPS across different sparse-view conditions.

Main Results:

• Spectral DPS achieved accurate water-bone separation and maintained uniform soft tissue appearance, even in out-of-distribution conditions.
• Decomposition remained robust down to 120 views, with minimal loss of structural detail and quantitative accuracy.
• At extreme sparsity (below 120 views), Spectral DPS showed increased variability and artifacts, but bony structures remained discernible down to 90 views.

Conclusions:

• Spectral DPS demonstrates significant potential for enabling accurate and robust material decomposition in sparse-view interventional CBCT.
• The framework's generalizability and single training requirement offer flexibility across different system configurations.
• The findings support the use of Spectral DPS for faster, lower-dose interventional imaging with improved material quantification.