Identification of glaucoma from fundus images using deep learning techniques

Area of Science:

• Ophthalmology
• Computer Science
• Artificial Intelligence

Background:

• Glaucoma is a leading cause of irreversible blindness globally, often stemming from elevated intraocular pressure.
• Early and accurate diagnosis of glaucoma is critical for preventing vision impairment.
• Manual glaucoma detection requires specialized expertise and significant experience.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN) algorithm for the automated diagnosis of glaucoma from fundus images.
• To assess the diagnostic performance of the proposed CNN model in classifying images as glaucomatous or normal.

Main Methods:

• A 13-layer CNN was trained on 1113 fundus images (660 normal, 453 glaucomatous) from multiple databases.
• Data augmentation increased the training set to 12012 images, with a 70/20/10 split for training, validation, and testing.
• The algorithm was implemented using Google Colab for accessibility and ease of use.

Main Results:

• The CNN model with a SoftMax classifier achieved 93.86% accuracy, 85.42% sensitivity, and 100% specificity and precision.
• A support vector machine (SVM) classifier integrated with the CNN model yielded superior results: 95.61% accuracy, 89.58% sensitivity, and 100% specificity and precision.

Conclusions:

• The deep learning model demonstrates significant capability in identifying glaucoma from retinal fundus images.
• The proposed system offers a potential tool for ophthalmologists to achieve faster, more accurate, and reliable glaucoma diagnoses.
• This automated approach can aid in the early detection and management of glaucoma, mitigating vision loss.