Adaptive Online Variational Bayesian Method based on Normal-Generalized Inverse Gaussian Prior for Bioluminescence...

Area of Science:

• Biomedical Imaging
• Optical Imaging
• Computational Biology

Background:

• Bioluminescence tomography (BLT) is a sensitive, non-invasive optical imaging technique for in vivo quantitative analysis.
• BLT reconstruction is challenged by the ill-posed nature of inverse problems, limiting spatial accuracy.
• Existing reconstruction methods struggle to balance signal sparsity and morphological preservation.

Purpose of the Study:

• To introduce an adaptive online variational Bayesian (AOVB-NGIG) method using a normal-generalized inverse Gaussian (N-GIG) prior for improved BLT reconstruction.
• To enhance the accuracy of spatial localization and morphological reconstruction in BLT.
• To provide a more robust foundation for preclinical and clinical pharmaceutical research.

Main Methods:

• Developed an AOVB-NGIG method incorporating an N-GIG prior to capture signal sparsity and complexity.
• Employed variational inference to maintain prior conjugacy and transform the optimization into a convex problem.
• Utilized the AdaDelta algorithm for adaptive learning rate adjustment to prevent local optima.

Main Results:

• Numerical simulations demonstrated superior performance of the AOVB-NGIG method over L1-TV, FISTA, and K-SVD.
• The proposed method achieved higher accuracy in tumor spatial localization.
• Enhanced morphological reconstruction of targeted regions was observed.

Conclusions:

• The AOVB-NGIG method significantly improves BLT reconstruction accuracy and morphological fidelity.
• This advancement offers a robust tool for preclinical and clinical pharmaceutical studies.
• The method provides a reliable foundation for future research in molecular imaging and drug development.