3D augmented fundus images for identifying glaucoma via transferred convolutional neural networks

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Glaucoma is a leading cause of irreversible blindness affecting millions worldwide.
• Current glaucoma diagnosis relies on subjective assessments like cup-to-disc ratio, leading to variability.
• Accurate and early detection of glaucoma is crucial to prevent vision loss.

Purpose of the Study:

• To investigate the efficacy of deep learning models utilizing augmented 3D optic nerve head (ONH) topographic maps for improved glaucoma diagnosis.
• To address the limitations of subjective analysis and inter-observer variability in current glaucoma screening methods.

Main Methods:

• Exploration of 3D optic nerve head (ONH) topography maps combined with deep learning (convolutional neural networks).
• Training of transferred AlexNet and VGG-16 networks using both 3D ONH topography and RGB fundus images.
• Comparison of diagnostic performance against models trained solely on 2D fundus images.

Main Results:

• 3D topographic maps of the ONH provided enhanced visualization of the optic cup and disc structures.
• Deep learning networks trained with the augmented 3D dataset achieved a diagnostic accuracy of 94.3%.
• This accuracy significantly surpassed that of networks trained using only 2D fundus images.

Conclusions:

• Deep learning models incorporating augmented 3D images enhance the accuracy of automated glaucoma detection from fundus images.
• This approach offers potential as an objective tool for computer-assisted diagnosis systems in glaucoma assessment.
• The use of 3D ONH data represents a significant advancement in objective glaucoma screening.