Material identification and quantification using MARS spectral CT.

I show, through extensions to the standard MARS material discrimination technique, that better use may be made of the available attenuation data than is presently achieved. MARS spectral CT measures the energy of each photon within a polychromatic x-ray beam as it passes through and is attenuated by an imaged object. This energy data, allocated to a five point scale, may be then used for both material identification and quantification. In this work, attenuation as a function of energy, spatial and contextural data are identified as separate information channels. Improved use of each information channel is demonstrated for a broad range of biomedical specimens, both by improved understanding of interactions during data acquisition and through extensions to the current linear least squares method. These extensions improve both material identification and material quantification, and extend material discrimination to combinations of materials which were previously not distinguishable. Machine learning is explored as an alternative approach to the linear least squares method which combines the strengths of the standard method and its extensions, with improved use of spatial and contextual information. Through proof-of-concept results, I demonstrate that the material identification problem is very well suited to machine learning. The method is applied to scans of human patients, incorporating knowledge of biological structures and materials to improve material identification. Incorporation of external information is particularly appropriate in clinical imaging, where assumptions may often be made about the materials, quantities, and structures imaged. Machine learning is expected to combine well with new and existing techniques for disease diagnosis and treatment monitoring. Thirdly, I demonstrate that the material identification tools developed for biomedical materials are useful for the study of small fossils. Scanning protocols are developed for material and structural investigation of such objects where a priori material knowledge may be limited or non-existent. Images are presented of a range of small plant, animal, vegetation and trace specimens which together demonstrate the utility of the MARS scanner for fossil imaging. The extent to which materials may be compared using standard attenuation signatures is explored using scanning electron microscopy for material identification. Although the wide variety of materials, structures and densities present makes material analysis of fossils particularly challenging, these images provide valuable structural information while also reducing common x-ray imaging artefacts. An automated method for analysis of fossil features is demonstrated