A Deep Learning Based Framework for Diabetic Retinopathy Detection EYEIVA

Project Title

A Deep Learning Based Framework for Diabetic Retinopathy Detection EYEIVA

Project Area of Specialization Software EngineeringProject Summary

Diabetic retinopathy is a global public health issue many patients with diabetic retinopathy are not treated in time because of inadequacies of the currently available screening programs. This project will classify diabetic retinopathy into five stages No diabetes retinopathy, mild diabetic retinopathy, Moderate diabetic retinopathy, Severe diabetic retinopathy, and proliferate diabetic retinopathy, after processing the image taken from the system through a deep learning algorithm (CNN). In this way, the presence of diabetic retinopathy in patients can easily be discovered automatically without conventional methods. The diabetic retinopathy detection system will be specially designed for hospitals, in which the doctors and opticians will be added by the administration of the hospital. This system will be a web-based computerized system that will help the doctors, and opticians detect diabetic retinopathy automatically by pre-processing the fundus image taken from the system (uploading the picture) using many elements such as CNN deep learning algorithm, pre-processing technique. The main advantage of this system will be that we can easily identify the retinal defect by simply uploading the fundus picture and getting reports indicating the stages of the disease. This system will detect diabetic retinopathy automatically which assists doctors in performing the task instead of manually checking each picture.

Project Objectives

The main objective of the system and our mission statement is to build a complete and accurate system for the detection of diabetic retinopathy in the attention eye.
The features of this system will be focused mainly on diabetic retinopathy detection using a fundus image and getting a report according to the fundus image. This system will be able to register admin of the hospital and further admin will be able to register doctors and opticians of the hospital. Admin can manage all the doctors and opticians registered on this system, doctor and optician will be able to upload fundus images image from the system and analyze the stages of the disease by uploading the image, within a few seconds he/she will be able to get report according to the fundus. So that doctors and opticians can share this report with patients by emailing them their reports with precautionary measures.

Project Implementation Method

Every software process model has some development phases, where the development team passes phases one by one and achieves their milestones. We are going to use the spiral development approach. The main principle of the spiral model is handling the risks.

• More effective,
• Focus on customer satisfaction,
• Rapid delivery,
• Flexible to change

It is a combination of waterfall and iterative models, in each phase in the spiral model begins with a design goal and ends with the client reviewing.

For the classification of the Diabetic retinopathy stages, we will use a deep learning algorithm, such as CNN, Alexnet, Densnet, mobile net, etc. Using the model, we will extract the features and it will be able to predict the stages of retinopathy from the images. The main steps of the spiral model are,

1. Planning phase: The very first stage of the spiral model development is where we will gather the requirements of the problem such as how the system will work and get the goal of the application such as by gathering the data and techniques.
2. Risk analysis: In this phase, we identify the risk in our project and make alternative solutions. We verify that the system we have implemented is bug-free by handing it over to the quality assurance team, who will test it and ensure that it is bug-free. When we found a risk, we suggest an alternative solution and implement them.
3. Engineering phase: This is the core development phase, in which we start coding and developing the product according to the requirements of the stakeholders after finalizing the tools and processes in this step. We begin the process of transforming the documented design into a working software application. After the completion of development, we start testing the product at the end of this phase.
4. Evaluation phase: In this phase, an evolution is done for the output of the product. The product is made available to customers for demonstration or practical use. Further iterations bring new features and fix faults to the already installed applications. If any changes are required, we will start the spiral and execute all phases again.

Phases:

1.In the first phase we will gather the dataset for training the model that we will use, and identify the features and milestones, we will analyze the architecture of the model and we will save the results for a better comparison between the results of different model and choosing one.

2.In this phase we will start developing the website frontend and backend, on Visual studio code in python and using the Django framework. The front end will include the Admin module, Doctor module, patient module, and optician module. In the backend, we will create a database for storing records that will also be implemented.

3.This phase will be more time-consuming and massive  according to the others because we will implement the above deep learning models and algorithms and test the accuracy of each algorithm on our dataset

Benefits of the Project

• Administration of the hospital will be able to manage doctors and opticians of the hospital
• The automated detection of diabetic retinopathy
• Classification of diabetic retinopathy up to 5 stages
• Report generation will be according to the stage of diseases detected.
• Sharing the report with the intended patient by emailing them the reports with precautionary measures.

Technical Details of Final Deliverable

The end product of "Eyeiva" will be a web-based application as well as a documentation manual containing all of the application's technical specifics. It will be built using Django, python, and the SQL database.

Final Deliverable of the Project Software SystemCore Industry HealthOther Industries Education , IT , Medical Core Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development Goals Life on LandRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	80000
Laptop	Equipment	1	50000	50000
MySQL Database hosting Subscription	Equipment	1	10000	10000
Googlecolab GPU	Equipment	1	10000	10000
printing	Miscellaneous	1	10000	10000