A State-of-the-Art Approach to Detect Diabetic Retinopathy Using Convolutional Neural Networks
CSEF · 2017 Mathematical Sciences
Overview
Objectives/Goals The purpose of this project was to determine if a state-of-the-art convolutional neural network image recognition algorithm could be developed to detect diabetic retinopathy in Fluorescein Angiography (fudus) photography to match the accuracy of a medical professional's mental model. Methods/Materials Materials included a Macbook with high computing power and internet access; a dataset of 35,000 retinal scans from Kaggle's diabetic retinopathy homepage; Inception v3 and ImageNet models; and machine learning libraries such as Tensorflow, Keras, scikit-learn, and matplotlib. A Java program was written to sort the scans based on their pre-classified labels. Python programs were written using the scikit-learn and matplotlib libraries to crop, reshape, denoise, visualize, and transpose the scans. The scans were split into a training set, with 29000 scans, a validation set, with 5000 scans, and a test set, with 200 scans. The Keras and Tensorflow libraries were used to build the Convolutional Neural Network architecture on top of the Inception base model and train the algorithm. The retinal scans in the test set were passed into the trained model and the algorithm's predictions were recorded for each image. The same set of scans was sent to three medical professionals. Results After the training process, a controlled test set of 200 scans was used to determine the accuracy, precision scores, and recall scores of the algorithm and the three medical professionals. The image recognition algorithm's accuracy was 79.50%, while the medical professionals' accuracies were 88.00%, 89.00%, and 88.00%. Conclusions/Discussion Although the computer algorithm's accuracy came close to the medical professionals' accuracies, it did not match them. The algorithm predicted scans without the disease more consistently than those with the disease. This algorithm breaks new grounds in this field by outperforming a prior model that was published in a Stanford PhD paper (trained on the same image dataset) by an 8% margin.
Summary statement
I trained a convolutional neural network to detect diabetic retinopathy to be almost as accurate as a medical professional and better than a prior model published in a Stanford PhD paper.
Help received
Dr. Kapil Kapoor (Vitreoretinal Surgeon), Dr. David Chia (Ophthalmologist), and Dr. Sanjay Kedhar (Gavin Herbert Eye Institute) classified the images in the test set. My father, Raj Neervannan, helped me understand the concepts required to complete the project.
Competition history
- CSEF 2017
Resources
Related projects
CSEF · 2018
A State-of-the-Art Approach to Detect Diabetic Retinopathy Using Convolutional Neural Networks and Adversarial Networks
CSEF · 2016
Automated Diagnosis of Diabetic Retinopathy in Fundus Images: A Novel Synthesis of Biological and Data Driven Approaches
ISEF · 2018
Developing an Algorithm for Detecting Diabetic Retinopathy from Retinal Images Using Machine Learning
ISEF · 2024
Detection of Diabetic Retinopathy With a Deep Learning Algorithm
ISEF · 2022
Comparing the Efficiency of Novel Point-Of-Care, Low-Cost Neural Networks in Identifying Specific Stages of Diabetic Retinopathy Across a Limited Retinal Dataset
ISEF · 2020
Detecting Common Retinal Diseases Through Use of Convolutional Neural Networks
AJAS · 2025
Automated Early Detection of Diabetic Retinopathy Using Deep Learning
ISEF · 2016
Automated Diagnosis of Diabetic Retinopathy Severity in Color Fundus Images Using a Novel Synthesis of Biological and Data Driven Approaches
Closest projects by meaning, across every fair and year in the corpus.
Browse more like this
Source: California Science & Engineering Fair public projects