Nonconvex regularizations in fluorescence molecular tomography for sparsity enhancement

In vivo fluorescence imaging has been a popular functional imaging modality in preclinical imaging. Near infrared probes used in fluorescence molecular tomography (FMT) are designed to localize in the targeted tissues, hence sparse solution to the FMT image reconstruction problem is preferred. Nonconvex regularization methods are reported to enhance sparsity in the fields of statistical learning, compressed sensing etc. We investigated such regularization methods in FMT for small animal imaging with numerical simulations and phantom experiments. We adopted a majorization-minimization algorithm for the iterative reconstruction process and compared the reconstructed images using our proposed nonconvex regularizations with those using the well known L-1 regularization. We found that the proposed nonconvex methods outperform L-1 regularization in accurately recovering sparse targets in FMT.