Classification of skin abnormalties using oblique incident diffuse reflectance spectroscopy

Due to the character of the original source materials and the nature of batch digitization, quality control issues may be present in this document. Please report any quality issues you encounter to digital@library.tamu.edu, referencing the URI of the item.Includes bibliographical references (leaves 67-72).Issued also on microfiche from Lange Micrographics.This thesis describes a study for the classification of skin abnormalities using oblique incident diffuse reflectance spectroscopy as part of the goal to provide a computer-assisted tool to dermatologists for lowering the number of benign biopsies. The first goal of this thesis was to improve the separability between the skin abnormalities. These skin abnormalities were divided by their melanocytic condition into two groups: (i) group 1 consisting of keratomas (benign), warts (benign), and carcinomas (malignant), and (ii) group 2 consisting of common nevi (benign), compound nevi (benign), junctional nevi (benign), against dysplastic nevi (pre-cancerous) and melanoma (cancerous). For each group a bootstrap based Bayes classifier was designed to separate benign from cancerous abnormalities. Genetic Algorithm was used to obtain the most effective set of image features for each classifier. A total of 584 images from 65 cases, 23 cases in group-1 and 42 cases in group-2, were used to train the classifiers. A blind study consisting of 222 images from 37 cases, 14 cases in group-1 and 23 cases in group-2 which were different from the ones used in the training part, was carried out to test the classifiers. The second goal of this thesis was to use the OIR data to compute the absorption spectra ([][]) and the reduced scattering spectra ([][]') and then use them to extract appropriate physiological parameters. This study was meant to provide an understanding of the physiological origins associated with different types of abnormalities