Mixture prior distributions and Bayesian models for robust radionuclide image processing

Area of Science:

• Medical Imaging
• Computational Science
• Statistical Modeling

Background:

• Radionuclide imaging is crucial for medical diagnosis and treatment planning.
• Combining imaging techniques improves accuracy but risks misalignment.
• Current image processing often uses global smoothing models, limiting adaptability.

Purpose of the Study:

• To develop a novel, locally adaptive image processing model for radionuclide imaging.
• To improve noise reduction and resolution in combined imaging techniques.
• To enhance diagnostic confidence and decision-making in nuclear medicine.

Main Methods:

• Proposed a Laplace and Gaussian mixture prior distribution for locally adaptive smoothing.
• Employed a fully Bayesian approach with multi-level hierarchical modeling.
• Utilized Markov chain Monte Carlo (MCMC) estimation for posterior distribution sampling.

Main Results:

• The novel model demonstrated superior noise reduction compared to existing methods.
• Image resolution was maintained without compromise.
• MCMC methods provided uncertainty quantification through posterior variance estimates.

Conclusions:

• Locally adaptive prior distributions offer a more realistic and robust modeling approach.
• The proposed Bayesian framework enhances reliability in nuclear medicine imaging.
• This methodology is applicable to various spatial inverse problems beyond medical imaging.