Spectral-Space Segmentation and Boundary Detection for Biomedical Hyperspectral Images

Biological Hyperspectral Imaging (BHSI) provides rich chemical and compositional information not regularly available from traditional imaging modalities. Retrieving segments and boundaries from the unstained tissue layers in such BHSI data without labels comes with a plethora of challenges that cannot be addressed directly from most high profile supervised artificial intelligence (AI) models. The lack of customized algorithms for the unique nature of our data requires the development of an open source platform to analyze this specific high-dimensional spatial and spectral data. In this thesis we develop a segmentation and boundary detection procedure using mathematical, statistical and machine learning algorithms that highlight complex boundaries and patterns that define segments of different tissue in BHSI data; especially, with a focus on the practical applications of analyzing eye tissue from lab mice suspended in a microgravity model devised to better understand the effect of Spaceflight-Associated Neuro-ocular Syndrome (SANS).