Joint estimation of tissue types and linear attenuation coefficients for photon counting CT

Area of Science:

• Medical Imaging
• Computational Imaging
• Radiology

Background:

• Photon counting detector CT (PCD-CT) offers enhanced tissue-type identification via material decomposition.
• Current iterative reconstruction methods often apply regularization based on image pixel values, limiting the use of projection data likelihood.
• Tissue-type identification is typically performed post-reconstruction, preventing integration with projection data probabilities.

Purpose of the Study:

• To develop a unified algorithm for image reconstruction and tissue-type identification in PCD-CT.
• To enhance both the quality of reconstructed CT images and the accuracy of tissue-type maps.
• To leverage projection data likelihood within a joint reconstruction and identification framework.

Main Methods:

• A novel framework employing maximum a posteriori (MAP) estimation with voxel-based latent variables for tissue types.
• Utilizing a voxel-based coupled Markov random field (MRF) to model organ continuity/discontinuity.
• Incorporating Gaussian distributions to capture statistical relationships between tissue types and attenuation characteristics.
• Quantitative comparison against filtered backprojection (FBP) and quadratic penalized likelihood (QPL) methods over 100 noise realizations.

Main Results:

• The proposed method demonstrated a superior noise-resolution trade-off compared to existing reconstruction techniques.
• Significant improvements in image quality metrics: bias reduced from -0.9 to -0.1 HU, and standard deviation (SD) decreased from 46.8 to 27.4 HU compared to QPL.
• Enhanced accuracy in tissue-type identification, improving from 80.1% with QPL to 86.9% with the proposed method.

Conclusions:

• The developed algorithm enables more accurate tissue-type identification in PCD-CT.
• Reconstructed CT images exhibit reduced noise and enhanced sharpness by integrating tissue-type information.
• This joint approach improves overall diagnostic quality in spectral CT imaging.