AI Framework for Glaucoma Screening Developed Using Portable Retinal Cameras in West Africa

Researchers have developed and evaluated an interpretable artificial intelligence (AI) framework designed for glaucoma screening in West Africa. This innovative system utilizes low-cost, portable handheld retinal cameras, offering a promising solution for early detection in areas with limited access to specialized medical equipment and ophthalmologists. The study specifically used data from 681 participants in Nigeria, encompassing a range of glaucoma statuses, to train and test the AI models.

The AI framework involves several component models for tasks such as vessel segmentation, cup and disc boundary segmentation, and optic nerve head feature extraction. These components are then combined into a final classification model capable of categorizing scans as glaucoma, glaucoma suspect, or non-glaucoma. A key aspect of this research is the emphasis on interpretability, using techniques like feature-weight analysis and Gradient-weighted Class Activation Mapping to provide clinicians with decision rationale visualizations.

The study compared the performance of the AI framework using images from the low-cost Volk Viva camera against a more expensive clinical tabletop camera (Canon CR-2-AF). Results showed that the Volk Viva, when combined with the AI, performed comparably well across various metrics, including segmentation and classification. This demonstrates that affordable, portable imaging devices, when augmented with sophisticated AI, can achieve diagnostic capabilities approaching those of more advanced and costly equipment.