DRIIS: Diabetic Retinopathy Analysis Based on Deep Learning Technique

Abstract

More than 85% of people with long-term diabetes are affected by Diabetic Retinopathy (DR), and it is a foremost reason for blindness in the 20–64 age range for both young and old patients. Non-proliferative Diabetic Retinopathy (NPDR) and Proliferative Diabetic Retinopathy (PDR) are two distinct kinds of DR. In the recent years, due to lack of precise diagnosis or timely prediction, 85% of people have lost their vision due to DR. Several techniques using diverse fundamental concepts were offered to diagnose this issue. According to the proposed method, the DR is categorized into five groups with a range of zero to four integers. Deep Convolutional Neural Network (DCNN) technique seems to work well enough, but describing the problem in a unique way to figure out how often the disease happens is still hard. One method proposed to automatically detect DR had 86.17% accuracy. This strategy utilized a CNN but lacked clinical training and validation data for the dataset. Even though CNNs have achieved remarkable results and are acclaimed for their generally high precise results in terms of image processing tasks, there are many hindrances and obstacles which affect the performance of CNN like the complex algorithm in terms of computation and processing time. To solve this problem of CNN, region proposals are identified, that can detect the region of interest based on the context or purpose. The next big problem is that there is no dataset of images of fundus that everyone or most people agree on. This makes it hard to use algorithms to analyze and get correct results. The Diabetic Retinopathy Image Database (DRiDB), for example, aims to get around this problem. So, our approach is to implement a Region Convolutional Neural Network (RCNN) for the detection of features. Usage of RCNN and commonly accepted database will ensure further accurate prediction of DR.