Related Experiment Video
Updated: Jun 24, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
Artificial intelligence based glaucoma and diabetic retinopathy detection using MATLAB - retrained AlexNet
Isaac Arias-Serrano1, Paolo A Velásquez-López1, Laura N Avila-Briones1
1School of Biological Sciences and Engineering, Universidad Yachay Tech, Urcuquí, Imbabura, 100119, Ecuador.
F1000Research
|June 3, 2024
Summary
Automated detection of glaucoma and diabetic retinopathy using convolutional neural networks (CNNs) shows high accuracy. Retinal image analysis with a retrained AlexNet CNN offers a promising tool for early disease identification.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Glaucoma and diabetic retinopathy (DR) are leading causes of irreversible blindness.
- Early detection via regular screening is crucial to prevent disease progression.
- Retinal fundus imaging is the primary diagnostic method for these conditions.
Purpose of the Study:
- To propose a MATLAB-retrained AlexNet convolutional neural network (CNN) for automated identification of glaucoma and diabetic retinopathy.
- To evaluate the performance of the proposed model against other CNN architectures.
- To incorporate Grad-CAM analysis for visualizing model predictions.
Main Methods:
- Retinal fundus images were used to train an AlexNet CNN via transfer learning.
- The model was trained to classify images into non-disease, glaucoma, and diabetic retinopathy categories.
- Benchmarking was performed using ResNet50 and GoogLeNet architectures.
Main Results:
- The retrained AlexNet CNN achieved a validation accuracy of 93.2% for classifying eye conditions.
- Benchmarking showed comparable performance with ResNet50 (93.8%) and GoogLeNet (90.4%).
- Validation accuracies for various configurations ranged from 89.7% to 94.3%.
Conclusions:
- A MATLAB-retrained AlexNet CNN effectively detects glaucoma and diabetic retinopathy from retinal images.
- Automated detection tools using CNNs are essential for early disease identification.
- CNNs offer accessible solutions to complement existing diagnostic technologies.
