Diffuse reflectance imaging modalities for characterization of highly scattering turbid media

Two different novel diffuse reflectance imaging modalities were studied as potential tools to characterize highly scattering turbid media. In the first case, diffuse reflectance polarization imaging (DRPI) was introduced to detect changes in optical properties in tissue-simulating phantoms. The changes in polarization patterns were correlated with the concentration of added molecules to the tissue phantoms. Radial and angular image analysis procedures were developed for these symmetric images to quantify variations in polarization patterns. This study showed that DRPI can be used as a tool to measure concentration changes of single compounds in turbid media. The findings of this basic research can be extended to real biological tissues to monitor the concentration changes of molecules, such as glucose, important for diabetic patients.; In the second study, oblique incidence diffuse reflectance spectroscopic imaging (DRSI) modality was employed to identify benign and cancerous skin lesions. In this imaging technique, multiple fibers were used to collect diffusely reflected light from the skin surface. The light collected by each fiber was separated into its spectral components with an imaging spectogram. Combining the 1-D spectral signals with multiple fibers provided a 2-D image that could then be analyzed using both spectrum as well as image analysis techniques, providing both frequency and spatial information. Unknown lesions were separated by developing image analysis, feature extraction, and classification techniques. For these images, texture features were investigated and new features introduced to test their success in separating the two classes. A region search algorithm was developed to find the optimum location of the region and width of region that produced features for best class separation. Over 95% correct classification rate was reached by the designed bootstrap-based Bayes classifier. Comparable results were obtained with the neural network classifier. 100% sensitivity and over 93% specificity was achieved. The results showed that the DRSI technique, when combined with feature extraction and classification algorithm, can be used to identify benign and cancerous skin lesions, and provide an alternative second opinion to the dermatologist on suspicious skin lesions in clinical settings.