Optical tomography, specically, diffuse optical tomography (DOT) and fluorescent molecular tomography (FMT) are promising functional imaging modalities with a high sensitivity and specicity. However, the inverse problem of DOT and FMT are ill-posed and ill-conditioned due to strong optical scattering in deep tissues, which results in poor spatial resolution for deep target imaging. It is well known that DOT and FMT image quality can be improved substantially by applying the structural guidance in the reconstruction algorithm.

In this dissertation, First, I conducted a feasibility study of computed tomography (CT) guided DOT system for breast cancer imaging. I built a noncontact projection style prototype DOT which consists of a laser at the wavelength of 650 nm and an electron multiplying charge coupled device (EMCCD) camera. We have validated the CT-guided DOT reconstruction algorithms with numerical simulations and phantom experiments, in which different imaging setup parameters, such as projection number of measurements and width of measurement patch, have been investigated.

Secondly, inspired by the kernel methods in machine learning, I introduced a kernel-based image reconstruction algorithm into anatomical image-guided DOT. Compared with conventional Laplacian approaches that include structural priors by regularization matrix, the developed method applied in this research incorporates a kernel matrix with the projection model into the objective function and does not require image segmentation. The optical absorption coefficient at each nite element node is represented as a function of a set of features obtained from

anatomical images such as computed tomography (CT) images. The proposed kernel method is validated with numerical simulations and agar phantom experiments. The proposed method utilized a CT volume data set without segmentation from a clinical breast CT system in the DOT.

Lastly, I implemented kernel-based anatomical guidance into the FMT image reconstruction. In FMT, the fluorophore concentration at each node is dened as a function of a set of feature vectors, which is directly extracted from the voxel intensities of the corresponding anatomical 3D images. This research studied the effects of voxel size and a number of nearest neighbors in the kernel method on the quality of reconstructed FMT images. The results indicate that the spatial resolution and the accuracy of the reconstructed FMT images have been improved substantially after applying the anatomical guidance with the proposed kernel method. The proposed method utilized magnetic resonance imaging (MRI) rat brain image in FMT simulation, which further proved that we do not need to segment the anatomical image for the kernel method. The proposed kernel method was found to be robust to the false positive guidance in the anatomical image.

As future work, the DOT prototype system will be integrated with a dedicated CT system, and clinical trials will be conducted using kernel-based image reconstruction algorithm.