Angle-Closure Detection in Anterior Segment OCT Based on Multilevel Deep Network.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Primary angle-closure glaucoma (PACG) is a leading cause of irreversible vision loss.
• Early detection of PACG is crucial for preventing vision impairment.
• Anterior segment optical coherence tomography (AS-OCT) is a key imaging modality for assessing the anterior eye segment.

Purpose of the Study:

• To develop and evaluate an automated deep learning system for detecting angle-closure glaucoma in AS-OCT images.
• To improve the accuracy and efficiency of PACG diagnosis.
• To leverage deep learning to identify subtle diagnostic cues in AS-OCT scans.

Main Methods:

• A multilevel deep network architecture was designed, incorporating three specific AS-OCT regions: global anterior segment, iris region, and anterior chamber angle (ACA) patch.
• A sliding window-based detector was employed for ACA region localization, treating detection as a regression task.
• Three parallel subnetworks extracted features from different regions, concatenated them, and fed them into a fully connected layer for classification.

Main Results:

• The proposed deep learning system demonstrated superior performance compared to existing detection methods.
• The system outperformed other deep learning approaches on two independent clinical AS-OCT datasets.
• The automated system effectively captured subtle visual cues indicative of angle closure.

Conclusions:

• The developed automated system offers a promising tool for accurate and efficient angle-closure glaucoma detection.
• Deep learning models can effectively analyze AS-OCT images for diagnosing PACG.
• This approach has the potential to aid clinicians in early PACG diagnosis and management.