Development of a deep residual learning algorithm to screen for glaucoma from fundus photography

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Glaucoma is a leading cause of irreversible blindness worldwide.
• Early detection and accurate diagnosis are crucial for effective glaucoma management.
• Fundus photography is a common method for visualizing the optic nerve head.

Purpose of the Study:

• To develop a deep residual learning algorithm for glaucoma screening using fundus photography.
• To evaluate the diagnostic performance of the developed algorithm against ophthalmology residents.

Main Methods:

• A deep learning algorithm, Deep Residual Learning for Image Recognition (ResNet), was developed using a training dataset of 1,364 fundus photographs with glaucoma and 1,768 without.
• The algorithm's diagnostic accuracy was validated on a testing dataset of 60 glaucoma patients and 50 normal subjects.
• Performance was measured using the area under the receiver operating characteristic curve (AROC) and compared to three ophthalmology residents.

Main Results:

• The ResNet algorithm achieved a significantly higher AROC of 96.5% (95% CI: 93.5 to 99.6%) in detecting glaucoma.
• The AROCs achieved by the three ophthalmology residents ranged from 72.6% to 91.2%.

Conclusions:

• Deep residual learning algorithms can achieve high diagnostic performance in screening for glaucoma from fundus photography.
• The developed ResNet algorithm outperformed ophthalmology residents in glaucoma detection accuracy.
• AI-powered tools show promise for enhancing glaucoma screening and diagnosis.