Adaptive k-sparse constrained dictionary learning strategy for bioluminescence tomography reconstruction.

Area of Science:

• Molecular imaging
• Biomedical research
• Medical physics

Background:

• Bioluminescence tomography (BLT) is a key molecular imaging technique.
• BLT reconstruction is often imprecise due to light scattering and ill-posed inverse problems.

Purpose of the Study:

• To develop an efficient and accurate reconstruction algorithm for BLT.
• To address the limitations of existing BLT methods in biomedical research.

Main Methods:

• Proposed an accelerated forward-backward splitting and difference of convex functions algorithm (AFBS-DCA) within a dictionary learning framework.
• Utilized k-sparsity for adaptive regularization parameter adjustment and generalized minimax-concave regularization for enhanced sparsity.
• Incorporated Nesterov's acceleration and DCA for efficient non-convex optimization and reduced computational complexity.

Main Results:

• AFBS-DCA achieved high reconstruction accuracy: 0.391 mm localization error, 0.774 Dice coefficient, and 0.872 contrast-to-noise ratio.
• Significantly reduced reconstruction errors compared to baseline methods (62.8%, 52.5%, 37.8%).
• Demonstrated superior performance in numerical simulations and experimental validation.

Conclusions:

• The AFBS-DCA method offers improved localization accuracy, morphological recovery, and robustness for BLT.
• This advancement holds potential for enhancing the practical applications of BLT in molecular imaging and biomedical research.