Generalized Gibbs priors based positron emission tomography reconstruction.

Area of Science:

• Medical Imaging
• Computational Imaging
• Bayesian Inference

Background:

• Image reconstruction is an ill-posed problem often requiring prior information for accurate results.
• Positron Emission Tomography (PET) imaging benefits from sharp edges to distinguish objects from background.
• Existing methods may struggle with noise suppression while preserving image details.

Purpose of the Study:

• To introduce a novel generalized Gibbs prior (GG-Prior) for enhanced image reconstruction.
• To leverage image affinity structure for improved reconstruction quality.
• To adapt Bayesian methods for better performance in PET imaging.

Main Methods:

• Developed a novel generalized Gibbs prior (GG-Prior) incorporating image affinity structure.
• Modified the paraboloidal surrogate coordinate ascent (PSCA) algorithm to integrate the GG-Prior.
• Implemented a local linearized scheme within the PSCA algorithm for GG-Prior incorporation.
• Tested the GG-Prior Maximum A Posteriori (MAP) reconstruction algorithm on simulated and real phantom data.

Main Results:

• The GG-Prior effectively suppresses noise while maintaining sharp image edges without oscillations.
• The proposed GG-Prior MAP reconstruction algorithm demonstrated superior performance compared to filtered backprojection (FBP) and Huber prior.
• Key improvements include significant noise reduction and enhanced edge preservation.
• Higher signal-to-noise ratio (SNR) conditions further highlighted the GG-Prior's advantages.

Conclusions:

• The GG-Prior is a powerful tool for image reconstruction, particularly in PET imaging.
• The proposed method offers a significant advancement in balancing noise suppression and edge preservation.
• GG-Prior based reconstruction provides higher quality images with improved diagnostic potential.