Diabetic retinopathy Detection and analysis

Project Title

Diabetic retinopathy Detection and analysis

Project Area of Specialization Artificial IntelligenceProject Summary

As of now, Doctors detect diabetic retinopathy through live screening of the retina using OCT Fundus image of retina is used by Ophthalmologists and Doctors for detecting diabetes.We are focusing on diabetes which is a deadliest disease effecting around 415 million people in the world.  46% of people with diabetes are undiagnosed.Ophthalmologists and Doctors use OCT machine for detecting the affected areas on retina.We are developing an application that will detect Diabetic Retinopathy and it’s stages by taking a fundus image of the retina as an input and provide the user with a result that will tell the user if they have the disease or not, and if they do, what stage is it at. We will make convolutional neural network model based on best medical performance measure to perform classification

Project Objectives

• Developing best diabetic retinopathy classification model

• Facilitating Doctors, Ophthalmologists, Nurses and all the normal users.

• Successfully deploying the trained model.

• Developing Models by using different methodologies of deep learning.

• Construct our own model based on previously developed best models.

Project Implementation Method

The softwares we are using are MATLAB, Android SDK. MATLAB is being used for image processing. We are using java language because Android applications are mostly developed in Java using the android studio and it has better support for android development than other languages and platform. XML is a markup language which will be used to design front end of our mobile   application. We will design our interface in XML language.Python provides an easy syntax which makes programming simple and is done in very compact manner. Main aim of Python programming language is to make the code readable. Some of main benefits of Python is Python provide a large standard library that manages memory automatically.  Because we have a large and well defined data set to learn from, this is a supervised learning problem and more specifically a classification problem. Our objective is to create accurate binary classification on our data; we wish to classify a retina image as having or not having various stages of diabetic retinopathy. There are 4 stages (1-4) and our first challenge was deciding how detect diabetes. We used convolutional neural networks, or CNNs, which are currently state of the art for image classification, as our learning model.

Benefits of the Project

The main benefit of this application is that, It can not only be used by professionals but a normal person can also use the application to detect the disease.

As of now, It is detected using Optical Coherence Tomography machine, These are not available everywhere because of it's expensive cost which makes it difficult for people to get examined. It increases the risk of the disease. In this era, everyone has a smart phone and people are familiar with it’s use. The huge benefit is that they can download the application free of cost, but to take a fundus image, they will be required to buy a 20D lens which costs from 24k-30k. We assure that the results would be as accurate as the one a doctor would give using an OCT machine.

Technical Details of Final Deliverable

A user is required to have an android smartphone with at least 4GB of RAM, and a 20D Fundus Lens to capture the retina image. The user can also use an already stored Fundus image in the phone to get results in the application. The application in designed in XML and all the functionalities are performed using Java. When the user first opens the application, they are greeted with a splash screen which leads to the Login/Signup window. New users are first required to register themselves on the application. Once they are registered, it is ready to use. The user can upload or capture the fundus image on the application for the detection of Diabetic Retinopathy. After photo is uploaded, It is ready to be processed. The uploaded photo is saved on Fire Base. The features are detected using MATLAB and afterwards, It is processed using CNN algorithms which results in accurate disease outcome. Once the processing is completed, the user is shown with the results. The result tells the user either they have Diabetic Retinopathy or not, and if they do, what stage is it. Furthermore, the user can save it’s result on the application for later use. 

Final Deliverable of the Project Software SystemCore Industry ITOther Industries Medical Core Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development GoalsRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	60000
Smartphone	Equipment	1	30000	30000
20D lens	Equipment	1	25000	25000
Internet	Equipment	1	5000	5000