Probabilistic atlas prior for CT image reconstruction.

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Reconstruction

Background:

• Statistical iterative reconstruction (SIR) offers superior image quality in computed tomography (CT) over traditional filtered backprojection (FBP).
• SIR methods benefit from effective noise modeling and the incorporation of prior information, driving continuous advancements.
• Low-dose CT imaging necessitates high-quality reconstruction from undersampled or noisy projection data.

Purpose of the Study:

• To introduce a novel prior information method for low-dose CT image reconstruction.
• To leverage probabilistic atlases for enhancing image quality in low-dose CT scans.

Main Methods:

• A two-phase approach involving a learning phase and a reconstruction phase.
• Learning phase: Construction of a 3D probabilistic atlas using a Laplacian mixture model and expectation maximization (EM) algorithm on a dataset of patient images.
• Reconstruction phase: Utilization of prior information from the probabilistic atlas to formulate the cost function for image reconstruction.

Main Results:

• Investigated low-dose imaging scenarios including reduced X-ray beam intensity and limited view angle data acquisition.
• Experimental studies with simulated and real chest screening CT data validated the approach.
• Demonstrated that the probabilistic atlas prior is a practically promising technique for low-dose CT imaging.

Conclusions:

• Prior information derived from a probabilistic atlas, built from diverse patient scans, is beneficial for low-dose CT imaging.
• The proposed method shows potential for improving diagnostic accuracy and patient safety in CT examinations.