Physics-informed and learning-based approaches to biomedical hyperspectral data analysis

Abstract

The high spectral and spatial resolution of hyperspectral imaging makes it a promising imaging technique for a wide range of biomedical applications. A recurring challenge is the handling and processing of the large amounts of data generated by the technique. The work of this thesis has focused on the analysis of an in vitro wound model dataset and a burn wound model dataset, utilizing supervised and unsupervised learning techniques and photon and heat transport modeling to extract information from the data. New insights on the characterizable optical property changes of these applications has been obtained, along with their relation to the tissue composition and underlying mechanisms. This enables development of targeted automated processing algorithms and better understanding of the technique.