Optical reflectance spectroscopy for cancer diagnosis : analysis and modeling
This dissertation focuses on the development of algorithms for analyzing and modeling of the signals from optical spectroscopy. This dissertation is motivated by the detection of oral cancer, but some of the methods developed can be generalized to epithelial cancers of other sites.
Two main topics are covered in this dissertation: Analysis and Modeling. For analysis, the focus is on developing algorithms to make diagnostic predictions. The analysis methods are empirically tested using an oral cancer dataset. Statistical analyses show that polarized reflectance spectroscopy has the potential to aid screening and diagnosis of oral cancer. Also, a novel adaptive windowing technique is developed to extract spectral features with fewer windows that retain the diagnostic information. For modeling, a Monte Carlo model simulating light-tissue interactions is presented to aid in the design of diagnostic instrumentation.