Ganglion cell layer analysis with deep learning in glaucoma diagnosis.

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Glaucoma is a leading cause of irreversible blindness worldwide.
• Early and accurate diagnosis is crucial for effective glaucoma management.
• Deep learning offers promising tools for analyzing complex medical imaging data.

Purpose of the Study:

• To evaluate and compare the diagnostic accuracy of two deep learning models (ResNet50 and VGG19) for glaucoma detection.
• To assess the models' performance using infrared images of the optic nerve, eye fundus, and ganglion cell layer (GCL).

Main Methods:

• A dataset of 498 eyes (312 glaucoma, 186 normal) was analyzed.
• Three infrared image types were used: fundus, optic nerve, and GCL, acquired via spectral-domain optical coherence tomography (SD-OCT).
• Deep learning models were developed on the MatLab platform using pre-trained ResNet50 and VGG19 neural networks.

Main Results:

• The deep learning models achieved high diagnostic precision: 96% for GCL images, 90% for optic nerve images, and 82-84% for fundus images.
• Area Under the Receiver Operating Characteristic Curve (ROC) values were highest for GCL images (0.96-0.97), followed by optic nerve (0.87-0.88) and fundus (0.79-0.81).
• Both ResNet50 and VGG19 models demonstrated comparable and strong performance, particularly with GCL imaging.

Conclusions:

• Deep learning models, especially when applied to GCL images, demonstrate high diagnostic precision, sensitivity, and specificity for glaucoma diagnosis.
• These findings suggest the significant potential of deep learning in improving glaucoma detection rates.
• The study validates the utility of infrared imaging combined with AI for ophthalmic diagnostics.