Glaucoma classification based on scanning laser ophthalmoscopic images using a deep learning ensemble method.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma diagnosis relies on clinical tests and imaging biomarkers like retinal nerve fiber layer (RNFL) thickness.
• Optic nerve head (ONH) en-face images from scanning laser ophthalmoscopy (SLO) offer valuable diagnostic information.
• Deep learning (DL) methods show potential for analyzing medical images but require robust performance in low-data scenarios.

Purpose of the Study:

• To evaluate the utility of ONH en-face SLO images for glaucoma detection.
• To demonstrate the effectiveness of a DL ensemble method for classifying glaucoma patients versus healthy controls, particularly in a low-data setting.
• To develop and benchmark a novel DL model for glaucoma classification using SLO images.

Main Methods:

• A dataset of 227 ONH en-face SLO images from 105 glaucoma patients and 122 controls was utilized.
• A custom convolutional neural network (CNN) architecture was designed for SLO image-based classification.
• The proposed DL ensemble method was compared against traditional machine learning classifiers using RNFL thickness and other feature-based methods.

Main Results:

• The DL ensemble model achieved a balanced accuracy of 0.962 in differentiating glaucoma patients from controls.
• The developed method demonstrated superior performance compared to all other tested classifiers, including those based on RNFL thickness.
• The study confirmed the potential of DL applied to SLO images for glaucoma diagnosis.

Conclusions:

• ONH en-face SLO images, analyzed with a DL ensemble method, provide a powerful tool for glaucoma detection.
• The proposed DL approach is effective even with a limited dataset, offering a viable diagnostic strategy.
• This study highlights the advancement of AI in ophthalmology for improved glaucoma patient management.