SABR-MIA: Support-driven Adaptive Bayesian Reweighting with Momentum Iterative Acceleration for Sparse Reconstruction in Bioluminescence Tomography.

Bioluminescence Tomography (BLT) is one kind of optical molecular imaging modality, whose principle is to estimate the three-dimensional spatial distribution of bioluminescent sources by using the optical information obtained from the surface of the imaging object. Due to the scattering effect of light and the limitation of detected surface photons, BLT has become an ill-posed problem, and the existing methods make it difficult to obtain satisfactory reconstruction results. To solve the sparse reconstruction problem of BLT, a support-driven adaptive Bayesian reweighted momentum iterative acceleration algorithm (SABR-MIA) is proposed to transform the optimization problem into a series of iteratively weighted subproblems. According to the "non-zero support" of the current iterative solution, the weights are adjusted dynamically, and the regularization weights of each component are updated based on Bayesian prior, thus effectively improving the reconstruction accuracy. Secondly, in the iterative process, to accelerate the convergence and enhance computational efficiency, this paper adopts the momentum iterative acceleration strategy (MIA). Specifically, based on the fast iterative shrinkage threshold algorithm (FISTA), adaptive threshold shrinkage is introduced, and the Nesterov acceleration strategy is combined to optimize gradient update. Simulation results confirm the method's effectiveness in reconstruction quality and accuracy.Clinical Relevance- The SABR-MIA method can significantly improve the quality of sparse reconstruction of BLT, which is conducive to further application in clinical studies such as cell tracking and tumor monitoring.