International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 05 Issue: 04 | Apr-2018 www.irjet.net p-ISSN: 2395-0072 © 2018, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 1355 Convolutional Neural Networks for Automatic Classification of Diabetic Retinopathy Dr. Sunil Bhutada 1 , Dr. Ch. Mukundha 2 , G. Shreya 3 , Ch. Lahari 4 1,2 Professor, Department of Information Technology, Sreenidhi Institute of Science and Technology, Telangana, India 3,4 Student, Department of Information Technology, Sreenidhi Institute of Science and Technology, Telangana, India ---------------------------------------------------------------------***--------------------------------------------------------------------- Abstract - : Diabetic Retinopathy (DR) is the main source of visual deficiency in diabetic patients. To tackle the problem of late diagnosis of diabetic retinopathy in affected people leading to permanent blindness, we are using convolutional neural networks for automated, quick and precise identification of the disease. The convolutional neural networks are trained by a labelled training dataset of fundoscopic images where they automatically learn the features of a diseased and normal retina and further use these features to detect the disease in the test fundoscopic images. Tensorflow’s Inception Model and its functions are used to construct the convolutional neural network and to train and test the neural network. This method of deep learning/machine learning to diagnose a DR is less timing consuming, almost accurate and can handle hundreds of test fundoscopic images at once, thereby helping a large number of people receive treatment in the earliest possible stages of the disease. Key Words: Convolutional Neural Networks, Diabetic Retinopathy, Supervised Learning, TensorFlow 1. INTRODUCTION Diabetic retinopathy (DR) is a medical condition in which damage occurs to the retina due to diabetes mellitus and is a leading cause of blindness, both partial and permanent. Diabetic retinopathy may cause no symptoms or only mellow vision issues. In long run, it can cause blindness. Without proper and timely detection, the disease may advance and lead to permanent blindness. Diabetic Retinopathy is characterized by the presence of hemorrhages on the retina and degree of severity is decided by the extent to which damage has been caused to the veins in the retina. Generally, Fundoscopic shots of the retina are screened by ophthalmologists to detect the disease. The idea is to use Machine Learning to detect and diagnose DR in a subject to accelerate the process in order to tackle the problem of doctor to diabetic patient ratios in India. A standout amongst the most widely recognized approaches to distinguish diabetic eye sickness is to have a pro look at photos of the back of the eye and decide if there are indications of the ailment, and assuming this is the case, how extreme it is. While yearly screening is prescribed for all patients with diabetes, numerous individuals live in regions without simple access to master mind. That implies a large number of individuals aren't getting the care they have to avoid loss of vision. We're energized by the outcomes, yet there's significantly more to do before a calculation like this can be utilized broadly. For instance, understanding of a 2D retinal photo is just a single step during the time spent diagnosing diabetic eye infection — now and again, specialists utilize a 3D imaging innovation to analyze different layers of a retina in detail. We are taking a shot at applying machine learning to that strategy. [1] Later on, these two integral strategies may be utilized together to help specialists in the analysis of a wide range of eye illnesses. To implement this idea, we are using an open source library for machine learning called Tensorflow. With the help of tensorflow, we can use the necessary functions required to construct a Convolutional Neural Network for image classification. 2. RELATED WORKS 2.1 Supervised Learning Supervised learning is the machine learning assignment of deducing a function from supervised training data. [2] The training data consists of a set of examples. In this, each example is a pair consisting of an input object and a desired output. A supervised learning algorithm analyzes the training data and results a function, which is referred as a classifier or a regression function. The function should predict the precise output value for any valid input data. This requires the taking in calculation to sum up from the preparation information to concealed circumstances sensibly. The parallel task in human and animal psychology is often referred to as concept learning. In supervised learning, the training data set should be well prepared to ensure the model works properly with accurate results. 2.2 Convolutional Neural Networks Convolutional Neural Networks are fundamentally the same as standard Neural Networks from the past: they are comprised of neurons that have learnable weights and predispositions. Every neuron gets a few sources of info, plays out a speck item and alternatively tails it with a non- linearity. The entire system still communicates a solitary differentiable score work: from the crude picture pixels toward one side to class scores at the other .[3] Despite everything they have a misfortune work (e.g. SVM/Softmax) on the last (completely associated) layer and every one of the tips/traps we produced for learning customary Neural