Regularization Based Reconstruction Algorithms For Fluorescence Molecular Tomography

Molecular imaging technique could observe the physiological activities in vivo bythe imaging methods at molecular and cellular levels in real time. In the imagingtechnique, the special molecule or cell is employed as the source of the image contrast,and the related changes of the molecules within the biological tissues could bevisualized. Molecular imaging technique has contributed a lot to the early detection ofdisease, targeted therapy, and drug development. Fluorescence molecular tomography(FMT) has become an important imaging tool of molecular imaging technique due toits high sensitivity, high temporal resolution, no ionizing, no radiation, low cost,convenient imaging process and so on. However, FMT has been applied mainly insmall animal imaging for its limited penetrating depth. Considering the absorption andscattering effect of light, how to reconstruct the three dimensional distribution offluorescent target inside the object with high accuracy and rapid speed from themeasured data on the surface, is always the key problem of FMT. The main work ofthis dissertation focuses on the reconstruction of fluorophore in FMT with continuouswave and is summarized as follows:1. A multilevel, hybrid regularization method is proposed to localize thefluorophore inside the object in fluorescence molecular tomography. Finite elementmethod (FEM) is utilized to numerically solve the problem of FMT. Generally, a finemesh is required for the reconstruction. Nevertheless, uniformly fine meshes wouldcreate a large number of unknown variables which might aggravate the ill-posedness ofFMT. Considering the effect of grid size, an adaptive finite element method isemployed for the reconstruction of the fluorophore. In this approach, the reconstructionregion is reduced from a coarse mesh to a finer mesh. In the first coarse mesh, sincethe region of the fluorescent target is much smaller than the reconstruction region, thesparsity regularization method is applied to find a solution that provides a good initialguess and guides the refinement of reconstruction region in the subsequent mesh levels.In the subsequent mesh levels, the reconstruction region is reduced largely since itincludes nodes with relatively large values. Landweber iterative regularization isperformed for the reconstruction instead of the sparisty regularization. The numericalsimulation experiments based on digital mouse and physical phantom experimentshave been conducted and the results have shown the potential feasibility of theproposed approach.2. In the quantitative fluorescent molecular tomography, a normalized Born approximation based two-stage reconstruction algorithm is developed. To reduce theinfluence of tissue heterogeneity, the normalized Born approximation is applied for themathematical model. The algorithm mainly includes two steps: the first step is tolocalize the fluorophore by l1-norm regularization method to fully take advantage ofthe sparsity of fluorophore. The first stage not only localizes the fluorophore, but alsoprovides the reconstruction region and the initial guess for the second stage. Then, inthe region where the fluorophore located, algebraic reconstruction technique (ART) isutilized to recover the fluorophore concentration. Numerical experiments and phantomexperiments have demonstrated the potential of the proposed method for quantitativefluorescent molecular tomography.3. We have made a comparative study between the reconstruction algorithmsbased on l1-norm and l2-norm for two imaging models of FMT. The first imagingmodel is adopted by most researchers where the fluorescent target is of small size tomimic small tissue with fluorescent substance, as exampled by the early detection oftumor. The second model is the reconstruction of distribution of the fluorescentsubstance in organs, which is essential to drug pharmacokinetics (PK). Apart fromnumerical experiments, in vivo experiments were conducted on a dual-modalityFMT/micro-computed tomography (CT) imaging system. The experimental results oflocation error and Full Width at Half Maximum (FWHM) of fluorescent yieldindicated that l1-norm regularization is more suitable for reconstructing the smallfluorescent target, while l2-norm regularization performs better for the reconstructionof the distribution of fluorescent substance.