Optical molecular imaging is a promising and cutting-edge technology with the features of high specificity, high resolution, and high sensitivity, aiming to quantitatively and qualitatively analyse in vivo life processes at cellular and molecular level. Fluorescence molecular tomography (FMT) is an important part of optical molecular imaging. Through fusing the fluorescence molecular imaging (FMI) and computer tomography to make the best use of each imaging modality, FMT can achieve the dynamic and consecutive observation of the three-dimensional biodistribution of fluorophore.
FMT is based on specialized fluorophore with specific cell targetting property, high qualified signal acquisition equipment like EMCCD, and through the mathematical model of light transmission among biological tissues to reconstruct the fluorephore biodistribution under excitation condition. However, due to the restriction on imaging processes, FMT faces high ill-posedness, high ill-conditionedness, as well as lack of robustness. Although conventional methods for FMT reconstruction is based on sparsity prior from the tumor biodistribution through adding L1-norm regularization to enhance the performance, introduce over-sparseness, spatial discontinuity, and low robustness. Hence, this study aims to overcome the above challenges, and proposes two novel reconstruction methods according to local spatial structural correlation and columns correlation in system matrix, to reduce the locating error, as well as improving the morphological similarity, relative sparsity, and robustness. The main contributions of my work can be summarized as follows.
1. Adaptive group orthogonal matching pursuit method (AGOMP)
In clinical applications, tumor spatial biodistribution inside imaging object not only satisfies the global sparsity, but is also accordance with local spatial structural correlation because of the clustering feature during the tumor growth. AGOMP method combines above two priors as group sparsity to design a novel local spatial structural regularization without the hard prior of tumor region from other modalities, and adopts sparsity adaptive orthogonal matching pursuit method to address the FMT reconstruction. The grouping strategy is based on small tetrahedron caused by finite element segmentation, and the grouped elements are packaged as a unit during searching and iteration. A series of numerical simulation experiments, based on digital mouse with both one and several tumors, were conducted, as well as in vivo mouse experiments. The results demonstrated the higher locating accuracy, more precise fluorescent yields, better morphological similarity, and more robust of AGOMP.
2. Regularized doubly orthogonal matching pursuit method (RDOMP)
Orthogonal matching pursuit method is based on greedy algorithm, through iteratively calculating residual error related element to update support set and complete FMT reconstruction. However, due to the impact of noise interference, segmentation deviation, and interaction among nodes on data acquisition and reconstruction, incorrected elements were always selected into support set and greatly affect the reconstruction performance. Therefore, this thesis proposed RDOMP method through synergistically integrating Gram-Schmidt (GS) orthogonalization with regularized orthogonal matching pursuit (ROMP) to decorrelate the elements in support set against remaining elements. Experiments based on the numerical mouse with double tumors and in vivo mouse were conducted to validate the enhancement of RDOMP. The reconstructed results demonstrated the better ability of atom selection compared with contrast methods, as well as avoid generating tiny tumor discontinuously located near the ground truth region, thus to enhance the performance of FMT reconstruction.