This Phase I research study will demonstrate the application of Artificial Neural Network (ANN) technology in the classification and grading of cataract images. The long-term objectives include the development and commercialization of a low-cost instrument for automatic diagnosis of cataracts. The major aim is to provide a standard and objective method to grade and classify different types of cataract. The cataract images are first feature extracted to produce descriptors that characterize the opacity severity and opacity relative shape. The grouping of opaque regions is based on the histogram uniformity. The relative shape analysis is based on the percentage of opacity in circular rings. These features will become the inputs to an Artificial Neural Network (ANN) using the back-propagation paradigm. The outputs of the ANN include the cataract gradings and types. The ANN classifier is first trained using a training set of cataract images of various types. After training, the performance of the ANN is evaluated using a test set of images and compared with the LOCSII or LOCSIII. The results in Phase I will become the foundation for further work in Phase II.
