CT Material Decomposition using Spectral Diffusion Posterior Sampling.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Science

Background:

• Spectral CT provides material-specific information beyond conventional CT.
• Accurate material decomposition is crucial for quantitative analysis in spectral CT.
• Existing model-based methods can be computationally intensive.

Purpose of the Study:

• To introduce a novel deep learning approach for material decomposition in spectral CT.
• To develop a faster and more stable variant of the diffusion posterior sampling (DPS) method.
• To evaluate the performance of the proposed method on different spectral CT systems.

Main Methods:

• Developed a deep learning approach combining unsupervised training priors with a physical measurement model.
• Introduced a "jumpstarted" process and gradient approximation for computational efficiency.
• Tested the method on dual-kVp and dual-layer detector spectral CT systems.

Main Results:

• The proposed DPS method achieved high accuracy (SSIM) using only 10% of iterations compared to model-based material decomposition (MBMD).
• Jumpstarted DPS (JSDPS) reduced computational time by over 85%.
• JSDPS demonstrated superior accuracy, lower uncertainty, and reduced computational cost compared to DPS and MBMD.

Conclusions:

• JSDPS offers a significant advancement in material decomposition for spectral CT.
• The method provides a fast and accurate solution for spectral CT data analysis.
• This deep learning approach holds great potential for clinical applications of spectral CT.