Accurate and efficient Methods for Fluorescence-mediated Tomography

• Applicant: Privatdozent Dr. Felix Gremse
• Subject Area: Medical Physics, Biomedical Technology ; Epidemiology and Medical Biometry/Statistics ; Computer Architecture, Embedded and Massively Parallel Systems
• Term: from 2017 to 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 380646556

Fluorescence-mediated tomography (FMT) enables the noninvasive and quantitative assessment of the three-dimensional fluorescence distribution in small objects, such as hands or lab mice. It is particularly suitable to investigate the biodistribution of potential therapeutic drugs, i.e. to determine the accumulation in targeted tissues but also to assess unwanted accumulation cites. Since the fluorescence signal does not decay exponentially with time, late imaging time points, e.g. after several days, are possible, which is relevant for many pharmacological studies. Furthermore, the expression of fluorescent proteins can be assessed and activatable tracers exist to determine the distribution of enzymes. In practice, FMT is particularly useful in combination with an anatomical modality, such as computed tomography, because the anatomical information is important for reconstruction, visualization and analysis.The fluorescence reconstruction is a mathematically and numerically challenging inverse problem which requires an accurate and efficient modeling of the light distribution inside the object. While this light distribution can be described accurately using the Boltzmann-equation, for numerical purposes strong approximate simplifications such as the diffusion equation are necessary. Furthermore, discretisation and assumptions about boundary conditions are necessary which affect computation time and reconstruction accuracy.In this project, GPU-accelerated Monte-Carlo-Simulations will be performed to generate artificial raw data to assess the influence of different modeling errors on the fluorescence reconstruction. Furthermore, the gain in accuracy by the use of higher order diffusion approximations will be investigated. To reduce the memory requirements for the reconstruction matrix and to enable usage of all available measurement data, an implicit matrix representation will be used. Furthermore, compatible iterative nonnegative linear solvers and methods to improve their convergence will be investigated. Furthermore, options to reconstruct absorption and scattering maps will be investigated. The implementation will utilize GPU-acceleration to profit from their high computational power. The simulated raw data will be used to assess improvements with respect to resolution, sensitivity and quantitativeness in silico. Additionally, the improvements will be validated with in vivo experiments. Fluorescent probes such as labeled antibodies will be used, which accumulate at known tissue regions. Furthermore, dual-labeled probes suitable for preclinical nuclear imaging devices will be used to assess the accuracy of the fluorescence reconstruction.

• DFG Programme: Research Grants