Detection of Diabetic Retinopathy (DR) Using Convolutional Neural Network (CNN) and Multiple Classifier Techniques in Machine Learning

Introduction

In this progressive world now day's diabetes has become a common disease that is affecting most of the world population. With diabetes many medical conditions are getting encountered that includes diabetic retinopathy which is mainly concerned with damage to retina that can cause vision loss. Detection of such diabetic eye in modern world can be achieved with help of machine learning techniques that include neural network, computer aided programs. In such systems fundus images are provided to the neural deep learning algorithms to extract features in the fundus images and then calculate the diabetic eye condition. Such mechanism helps us detect the conditions in fundus images more accurately and efficiently. Diabetic retinopathy has mainly 2 types that are to be detected for retinal diseases like Non-Proliferative Diabetic Retinopathy disease(NPDR) and Proliferative Diabetic Retinopathy disease(PDR)(Wang et al., 2020).Medical industry nowadays has advanced in a manner where high end technologies are used for early detection and analysis of diseases that are hard to encounter with normal procedure of medical field. One of such disease is Diabetic retinopathy(DR) further classifield as Non-Proliferative Diabetic Retinopathy (NPDR) disease and Proliferative Diabetic Retinopathy disease(PDR) conditions which are caused by diabetes. Early detection of NPDR is a challenging task and it requires examination of fundus images in a amplified manner. To overcome these early detection of DR we are proposing a automated system that will be using machine learning classifier techniques with combination of convolutional neural network(CNN) (Alves, 2020) to self train the system on its own and detect the early stages of retinal scans by feature extraction and use of existing retinal scan databases. Hence the system will eliminate the human flaw of inability to detect early DR in diagnosis and will help us treat the patient in early stages itself. CLAHE method will be used for preprocessing of fundus images which will be used by CNN. CNN will extract features from the input set of fundus images and pass it to classifiers where the accurate result will be obtained to detect the DR. Medical industry nowadays has advanced in a manner where high end technologies are used for early detection and analysis of diseases that are hard to encounter with normal procedure of medical field. One of such disease is Diabetic retinopathy(DR) further classified as Non-Proliferative Diabetic Retinopathy (NPDR) and Proliferative Diabetic Retinopathy(PDR) conditions which are caused by diabetes. Early detection of NPDR is a challenging task and it requires examination of fundus images in a amplified manner. To overcome these early detection of DR we are proposing a automated system that will be using machine learning classifier techniques with combination of convolutional neural network(CNN) to self train the system on its own and detect the early stages of retinal scans by feature extraction and use of existing retinal scan databases. Hence the system will eliminate the human flaw of inability to detect early DR in diagnosis and will help us treat the patient in early stages itself. CLAHE (Lands et al., 2020) method will be used for preprocessing of fundus images which will be used by CNN. CNN will extract features from the input set of fundus images and pass it to classifiers where the accurate result will be obtained to detect the DR.