Robust transformed l1 metric for fluorescence molecular tomography.
Yating Yuan,Huangjian Yi,Dizhen Kang,Jingjing Yu,Hongbo Guo,Xuelei He,Xiaowei He
DOI: https://doi.org/10.1016/j.cmpb.2023.107503
IF: 6.1
2023-01-01
Computer Methods and Programs in Biomedicine
Abstract:BACKGROUND AND OBJECTIVE:Fluorescence molecular tomography (FMT) is a non-invasive molecular imaging modality that can be used to observe the three-dimensional distribution of fluorescent probes in vivo. FMT is a promising imaging technique in clinical and preclinical research that has attracted significant attention. Numerous regularization based reconstruction algorithms have been proposed. However, traditional algorithms that use the squared l2-norm distance usually exaggerate the influence of noise and measurement and calculation errors, and their robustness cannot be guaranteed.
METHODS:In this study, we propose a novel robust transformed l1 (TL1) metric that interpolates l0 and l1 norms through a nonnegative parameter Ī±ā(0,+ā). The TL1 metric looks like the lp-norm with pā(0,1). These are markedly different because TL1 metric has two properties, boundedness and Lipschitz-continuity, which make the TL1 criterion suitable distance metric, particularly for robustness, owing to its stronger noise suppression. Subsequently, we apply the proposed metric to FMT and build a robust model to reduce the influence of noise. The nonconvexity of the proposed model made direct optimization difficult, and a continuous optimization method was developed to solve the model. The problem was converted into a difference in convex programming problem for the TL1 metric (DCATL1), and the corresponding algorithm converged linearly.
RESULTS:Various numerical simulations and in vivo bead-implanted mouse experiments were conducted to verify the performance of the proposed method. The experimental results show that the DCATL1 algorithm is more robust than the state-of-the-art approaches and achieves better source localization and morphology recovery.
CONCLUSIONS:The in vivo experiments showed that DCATL1 can be used to visualize the distribution of fluorescent probes inside biological tissues and promote preclinical application in small animals, demonstrating the feasibility and effectiveness of the proposed method.