EyeSpy Diagnosis: Developing a Smartphone-Based Non-Invasive Intelligent System for Detecting Eye Fundus Anomalies via Machine Learning

Overview

Diabetic retinopathy (DR) is the leading cause of preventable blindness in the world. Due to the necessity of expensive ophthalmoscopic equipment and a trained professional, the screening process is inaccessible to millions of people. The goal of this work was to develop an end-to-end solution for triaging DR in an accessible, non- invasive, timely, and affordable manner. First, a machine learning (ML) model was developed for image analysis using convolutional neural networks and feature extraction with the Keras framework in Python. The model was trained on the EyePACS open-source, annotated dataset to ensure its validity over a wide range of fundus images. Second, programs were created to segment blood vessels and extract microaneurysms from the image. Third, a smartphone application was programmed to take images and house the ML model. Finally, a versatile smartphone lens attachment was 3D printed to seamlessly integrate with the application and provide testable DR classification results within minutes. The ML model was validated on the EyePACS testing dataset and obtained statistically comparable results to those of an ophthalmologist in a clinical setting. The applications of this project are twofold: first, this presents a novel solution to grading the severity of DR and identifying fundus anomalies, and second, this approach can be easily adapted to diagnose other retina-related conditions. The project was deemed successful as the system is an intuitive and affordable tool for accurately diagnosing DR and has the potential to mitigate the vision impairment issues currently faced by 18% of the global population.

Competition history

• JSHS 2020 Category not listed

Resources

• National JSHS abstract book ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.