Automated Detection and Explanation of Diabetic Retinopathy

Introduction

3rd Year Dissertation Project

Diabetic Retinopathy is a commmon complication of diabetes which affects the sight. It is caused by damage to the blood vessels of the light-sensitive tissue at the back of the eye (retina). Due to a large number of small symptoms it is optimal for computer vision techniques.

Methodology

The project had 3 sections.

Detection of Diabetic Retinopathy (DR)

This involved building a CNN (InceptionResNetV2) pre-trained on imagenet, and utilising transfer learning and fine-tuning to train the model to classify the input image into a Diabetic Retinopathy severity level.

Exaplainable AI

This involved using Grad-CAM++ to generate heatmaps of the input image to show which parts of the image the model was focusing on when making its prediction. GradCAM++ was a novel approach for the explainability of DR detection with CNNs, and provided and better output that its pre-decessor GradCAM.

Production

The final stage was actually deploying the trained deep learning model into production. This involved bulding a simple Python application for Opthalmologists to use in real-life, which implemented both the InceptionResNetV2 and GradCAM++.

Key Technologies

• Python
• Tensorflow

Links

• Github