A filtered backprojection MAP algorithm with nonuniform sampling and noise modeling.

Area of Science:

• Medical imaging
• Computational imaging
• Image reconstruction algorithms

Background:

• Iterative algorithms like Landweber and MAP are standard for image reconstruction.
• Filtered Backprojection (FBP) is a faster, though often less robust, alternative.
• Existing FBP methods struggle with non-uniform sampling and accurate noise modeling.

Purpose of the Study:

• To extend the filtered backprojection (FBP) algorithm to incorporate characteristics of iterative Maximum a Posteriori (MAP) algorithms.
• To develop an FBP algorithm capable of modeling projection noise using a view-based weighting scheme.
• To enable FBP reconstruction with non-uniformly sampled projection data.

Main Methods:

• Developed a new objective function for FBP incorporating view-based noise weighting and a Bayesian prior term.
• Derived a frequency-domain window function for each iteration, modifying the ramp filter and windowing function.
• The algorithm accommodates non-uniform sampling through modified backprojection weighting.

Main Results:

• Computer simulations demonstrated that the new FBP-MAP algorithm yields reconstructions comparable in resolution and noise texture to iterative Landweber MAP.
• The noise modeling effectively reduced streaking artifacts in transmission x-ray CT, particularly in low-dose scenarios.
• The algorithm successfully processed non-uniformly sampled projection data.

Conclusions:

• A view-based noise weighting scheme can be integrated into the FBP algorithm's window function.
• The enhanced FBP algorithm achieves results similar to iterative MAP algorithms with modifications to the ramp filter.
• The FBP-MAP algorithm effectively addresses challenges of non-uniform sampling and sensitivity in image reconstruction.