Diagnosis of Diabetic Retinopathy Using Principal Component Analysis (PCA)

Diabetic retinopathy is an eye disease due to diabetes, which is not detected in its early stage, may cause vision loss. The need of automated diagnosis methods of diabetic retinopathy increases day by day because of its severity. Authors proposed the design of diabetic retinopathy automated diagnosis system based on neural networks. Multi Layer Perceptron (MLP), Principal Component Analysis (PCA), Generalized Feed Forward (GFF) neural networks are employed to design automated classifier system in first experiment. In second experiment, the input dimensionality reduction method based MLP, GFF neural networks classifier systems are designed and compared the performances. In experiment 1, the average classification accuracy for MLP network is nearly 99.00% whereas GFF-NN has 92.00% on CV data. In experiment 2, using Principal Components (PCs), the average classification accuracy for MLP network is nearly 97.22% whereas GFF-NN has 84.37% on CV data. The N/P ratio for MLP and GFF networks is large in second experiment which is 0.273 and 0.219 respectively having less neural network's architecture complexity.