Compensation for nonuniform resolution using penalized-likelihood reconstruction in space-variant imaging systems.

Area of Science:

• Medical Imaging
• Image Reconstruction
• Statistical Modeling

Background:

• Statistical imaging systems often produce images with nonuniform resolution.
• This non-uniformity, characterized by space-variant and/or anisotropic responses, impacts image quality.
• Previous work developed a space-variant penalty for penalized-likelihood (PL) reconstruction to achieve uniform resolution in idealized systems.

Purpose of the Study:

• To demonstrate the efficient calculation and application of a space-variant penalty to complex, space-variant imaging systems.
• To evaluate the penalty's effectiveness in improving resolution uniformity for large field-of-view PET and 2D SPECT systems.
• To compare the proposed PL approach with other reconstruction methods.

Main Methods:

• Developed and applied an efficient calculation method for a space-variant penalty in PL reconstruction.
• Tested the method on a large field-of-view PET system with space-variant geometric response due to crystal penetration.
• Applied the penalty to a 2D SPECT system with depth-dependent Gaussian detector responses.

Main Results:

• The proposed PL approach demonstrated improved relative resolution uniformity compared to other reconstruction methods.
• Noise performance was similar between the PL and post-smoothed maximum-likelihood (ML) approaches when resolution was matched.
• Tradeoffs in resolution uniformity were observed between the PL and post-smoothed ML methods depending on desired resolution.

Conclusions:

• The space-variant penalty offers a viable method for achieving nearly uniform resolution in complex imaging systems.
• The choice between PL and post-smoothed ML may depend on computational factors and desired resolution characteristics.
• A hybrid approach combining penalty and post-smoothing is presented as a potential method to optimize resolution uniformity.