SMARTPHONE BASED SYSTEM FOR DIABETIC RETINOPATHY DETECTION USING ARTIFICIAL NEURAL NETWORK

Early diagnosis of diabetic retinopathy for the treatment of the disease has been failing  to reach diabetic people living in rural areas. The shortage of trained ophthalmologists,  limited availability of healthcare centers, and expe

Project Title

Project Area of Specialization

Artificial Intelligence

Project Summary

Early diagnosis of diabetic retinopathy for the treatment of the disease has been failing

to reach diabetic people living in rural areas. The shortage of trained ophthalmologists,

limited availability of healthcare centers, and expensiveness of

diagnostic equipment are among the reasons.

In this project we are going to develop a Deep Learning Model of Diabetic Retinopathy

with an android application. First the Dataset of Diabetic Retinopathy is trained after

that through mobile phone we attach a handheld ophthalmoscope and take high resolution

Fundus images and then Deep learning model then detect and classify the

Diabetic Retinopathy and grade the stages.

Project Objectives

To Develop an artificial intelligence (AI)-based automated software for detection of diabetic retinopathy (DR) by fundus images using a smartphone.

Identify and classify different Stages of Diabetic Retinopathy
Train and test a model in TENSORFLOW (ML/DL).
Acquire images taken from a 20D Retina lens attached to a smartphone for testing and training of models.
Develop a user-friendly Android App.
Minimize chances of False-Diagnosis.
Reduce the amount of time and cost required for diseases diagnosed using a Microscope.
To provide people living in remote areas a quick and easy method for Detection of Diabetic Retinopathy without the need of an ophthalmologist.( eye specialist )

Project Implementation Method

Artificial Neural Network
Image Processing
For training a deep learning model, a dataset of fundus images from Kaggle website will

be used. The dataset contains fundus images labeled into five stages of DR.

Image Pre-processing For model training, 2000 images were selected from each stage of DR.

In order to exploit high performance from a deep learning model and smaller training dataset;

image preprocessing will be performed. after that large datasets will be used.

The fundus images are then cropped to remove unnecessary black background pixels.
The local average was subtracted from the cropped fundus image.

The designed deep learning model will be imported to the smartphone.

fig. 2 Project Implementation Flowchart

Benefits of the Project

Early Detection of Diabetic Retinopathy.
This mobile eye examination system is easily accessible to anyone living in the city or

in a village.

Cost-efficiency: Existing methods include costs incurred for expensive sophisticated fundus

cameras and operating technicians. Our system, on the other hand, uses a low-cost portable

handheld 20D retina lens and a smart-phone.

Portability: Our compact system is easy-to-deploy on field locations that are hundreds of

miles away from specialists. This is unlike present-day heavy equipment.

Ease of operation: Presently, ophthalmologists rely extensively on specially trained

personnel to capture images using the fundus cameras. Our system simplifies the task of

image collection independent of the operator.

Decision-making capability: In contrast to other systems which require interpretation of

images by an ophthalmologist, our system provides a first-hand assessment of conditions to

general practitioners and emergency room physicians

Technical Details of Final Deliverable

The final deliverable will be an Android App which will acquire Fundus images from a 20D lens attached to a smartphone. The App will detect and grade the stages of Diabetic Retinopathy.