The design and application of ICAT : Interactive Cluster Analysis Toolkit

An individual afflicted with diabetes mellitus can develop a condition in which lipids and blood accumulate in the eye. Such deposits are indicative of the early stages of diabetes and potential blindness. In diabetic retinopathy, early de tection of these anomalies can aid in diagnosis and prevention. Using cluster analysis on retinal images, computer-generated statistics of cluster size, radius of gyration, ellipsoidal eccentricity, and average texture density can be used to classify such anomalies. In lieu of an automated approach, this thesis promotes a general framework to assist a physician in the detection and classification of reti-nal abnormalities. The proposed Interactive Cluster Analysis Toolkit (or ICAT) utilizes the Enhanced Hoshen-Kopelman algorithm to provide a highly adaptable method for cluster analysis. Within the context of diabetic retinopathy, different neighborhood rules implemented within ICAT may provide better approaches for classifying retinal features such as neovascularization and exudates. The flexible design of ICAT allows new metrics for characterizing cluster geometry or new neighborhood rules for cluster identification to be easily incorporated.