Sparse-graph manifold learning method for bioluminescence tomography.

Area of Science:

• Biomedical Imaging
• Optical Molecular Imaging
• Preclinical Research

Background:

• Bioluminescence tomography (BLT) is crucial for preclinical tumor imaging, monitoring, and therapy guidance.
• Accurate tumor spatial localization and morphology analysis are critical challenges in BLT.
• Existing BLT reconstruction methods often focus on either sparsity or morphology, lacking versatility.

Purpose of the Study:

• To develop a versatile algorithm for simultaneous tumor spatial localization and morphology recovery in BLT.
• To address the limitations of existing BLT reconstruction methods.

Main Methods:

• Proposed a Sparse-Graph Manifold Learning (SGML) method integrating non-convex sparsity and dynamic Laplacian graph models.
• Developed an iteratively reweighted soft thresholding algorithm (IRSTA) for solving the SGML model based on nonconvex optimization.
• Validated the method through numerical simulations and in vivo experiments.

Main Results:

• The SGML method demonstrated superior performance in spatial localization and tumor morphology recovery compared to existing methods.
• The algorithm effectively balanced source sparseness and morphology.
• Results were consistent across various source settings in both simulations and in vivo studies.

Conclusions:

• The proposed SGML method offers a versatile solution for simultaneous spatial localization and morphology recovery in BLT.
• The SGML method shows significant potential for applications in optical tomography and optical molecular tomography.
• This advancement facilitates improved preclinical tumor imaging and related therapeutic applications.