Advanced tomographic image reconstruction algorithms for Diffuse Optical Imaging

Diffuse Optical Imaging is relatively new set of imaging modality that use infrared and near infrared light to characterize the optical properties of biological tissue. The technology used is less expensive than other imaging modalities such as X-ray mammography, it is portable and can be used to monitor brain activation and cancer diagnosis, besides to aid to other imaging modalities and therapy treatments in the characterization of diseased tissue, i. e. X-ray, Magnetic Resonance Imaging and Radio Frequency Ablation. Due the optical properties of biological tissue near-infrared light is highly scattered, as a consequence, a limited amount of light is propagated thus making the image reconstruction process very challenging. Typically, diffuse optical image reconstructions require from several minutes to hours to produce an accurate image from the interaction of the photons and the chormophores of the studied medium. To this day, this limitation is still under investigation and there are several approaches that are close to the real-time image reconstruction operation. Diffuse Optical Imaging includes a variety of techniques such as functional Near-Infrared Spectroscopy (fNIRS), Diffuse Optical Tomography (DOT), Fluorescence Diffuse Optical Tomography (FDOT) and Spatial Frequency Domain imaging (SFDI). These emerging image reconstruction modalities aim to become routine modalities for clinical applications. Each technique presents their own advantages and limitations, but they have been successfully used in clinical trials such as brain activation analysis and breast cancer diagnosis by mapping the response of the vascularity within the tissue through the use of models that relate the interaction between the tissue and the path followed by the photons. One way to perform the image reconstruction process is by separating it in two stages: the forward problem and the inverse problem; the former is used to describe light propagation inside a medium and the latter is related to the reconstruction of the spatio-temporal distribution of the photons through the tissue. Iterative methods are used to solve both problems but the intrinsic complexity of photon transport in biological tissue makes the problem time-consuming and computationally expensive. The aim of this research is to apply a fast-forward solver based on reduced order models to Fluorescence Diffuse Optical Tomography and Spatial Frequency Domain Imaging to contribute to these modalities in their application of clinical trials. Previous work showed the capabilities of the reduced order models for real-time reconstruction of the absorption parameters in the brain of mice. Results demonstrated insignificant loss of quantitative and qualitative accuracy and the reconstruction was performed in a fraction of the time normally required on this kind of studies. The forward models proposed in this work, offer the capability to run three-dimensional image reconstructions in CPU-based computational systems in a fraction of the time required by image reconstructions methods that use meshes generated using the Finite Element Method. In the case of SFMI, the proposed approach is fused with the approach of the virtual sensor for CCD cameras to reduce the computational burden and to generate a three-dimensional map of the distribution of tissue optical properties. In this work, the use case of FDOT focused on the thorax of a mouse model with tumors in the lungs as the medium under investigation. The mouse model was studied under two- and three- dimension conditions. The two-dimensional case is presented to explain the process of creating the Reduced-Order Models. In this case, there is not a significant improvement in the reconstruction considering NIRFAST as the reference. The proposed approach reduced the reconstruction time to a quarter of the time required by NIRFAST, but the last one performed it in a couple of seconds. In contrast, the three-dimensional case exploited the capabilities of the Reduced-Order Models by reducing the time of the reconstruction from a couple of hours to several seconds, thus allowing a closer real-time reconstruction of the fluorescent properties of the interrogated medium. In the case of Spatial Frequency Domain Imaging, the use case considered a three-dimensional section of a human head that is analysed using a CCD camera and a spatially modulated light source that illuminates the mentioned head section. Using the principle of the virtual sensor, different regions of the CCD camera are clustered and then Reduced Order Models are generated to perform the image reconstruction of the absorption distribution in a fraction of the time required by the algorithm implemented on NIRFAST. The ultimate goal of this research is to contribute to the field of Diffuse Optical Imaging and propose an alternative solution to be used in the reconstruction process to those models already used in three-dimensional reconstructions of Fluorescence Diffuse Optical Tomography and Spatial Frequency Domain Imaging, thus offering the possibility to continuously monitor tissue obtaining results in a matter of seconds