Abstract
Motivated by glaucoma's status as the second leading cause of blindness, the increasing numbers of glaucoma patients, the difficulty of early glaucoma detection, and the disease's irreversibility, we propose strategies for constructing tools such as automated classifiers or fundus image glaucoma-defects enhancers to effectively differentiate between normal and glaucomatous eyes. We first consider the commonly used eye characteristic, macular thickness. We propose an effective automated classifier (named BPNN+GA), which is a backpropagation neural network with an initial value obtained via a GA algorithm. The input data for BPNN+GA are functions of outer and inner superior and inferior macular thickness. The proposed algorithm shows that the performance measures, sensitivity and specificity, are both improved. Specifically, sensitivity and specificity are 94.2% and 99.9%, respectively. Moreover, we also propose a new fundus image glaucoma-defects enhancer using another type of eye characteristic, the retinal nerve fiber layer (RNFL) defect. We have shown that the RNFL defect can be enhanced by applying the multiscale retinex with color restoration (MSRCR) technique to the data from the fundus image rather than the data on macular thickness.