ELECTRONIC HEALTH RECORD OF DIABETIC RETINOPATHY IMAGE CLASSIFICATION BY FEATURE OPTIMIZATION AND RNN MODEL

Abstract

 Diabetic retinopathy (DR) is a widespread and severe complication of diabetes mellitus that can ultimately result in vision loss or blindness. Precise identification of the different stages of DR is crucial for ensuring early diagnosis and appropriate treatment. In this study, a novel approach is proposed to enhance DR classification by accurately detecting retinal lesions in fundus images. This paper has proposed Elephant Herd Based Diabetic Retinopathy Image Classification (EHDRIC) model that optimize input feature image by transforming spatial values into discrete cosine frequency feature. Transformed feature is optimized by elephant herd algorithm. Extracted Effective coefficient features of images were used for the distribution feature. This distribution reduces the dimension of the input feature for the training of the recurrent neural network. Experiment was done on real dataset of diabetic retinopathy images and result shows that proposed EHDRIC has improved the correct class detection accuracy.