Predicting Visual Field Loss in Glaucoma Using OCT and Deep Learning: A Comparative Study of U-Net Variants

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Glaucoma is a progressive optic nerve disease causing irreversible visual field loss.
• Optical coherence tomography (OCT) and visual field (VF) tests are crucial for monitoring glaucoma progression.
• Deep learning models offer potential for predicting functional outcomes from structural OCT data.

Purpose of the Study:

• To evaluate the efficacy of three deep learning models (R2 U-Net, Dense U-Net, Nested U-Net) in predicting visual field outcomes.
• To utilize retinal nerve fiber layer (RNFL) thickness maps from OCT images as input for these predictive models.
• To assess the performance of generative AI in predicting glaucoma-related visual field loss.

Main Methods:

• A retrospective study analyzed OCT and VF data from 1640 glaucoma patients.
• Three deep learning architectures (R2 U-Net, Dense U-Net, Nested U-Net) were trained to predict VF outcomes from RNFL thickness maps.
• Model performance was evaluated using metrics including MSE, MAE, PSNR, and SSIM.

Main Results:

• R2 U-Net demonstrated superior performance, achieving the lowest MSE and MAE, and highest SSIM and PSNR.
• Nested U-Net and Dense U-Net showed lower predictive accuracy compared to R2 U-Net.
• The study confirmed R2 U-Net's capability for accurate VF prediction from OCT data.

Conclusions:

• Generative AI models, particularly R2 U-Net, show significant potential for predicting visual field loss in glaucoma.
• These findings suggest that AI can enhance glaucoma diagnosis and aid in personalized treatment strategies.
• The study pioneers the application of generative AI for predicting VF loss using OCT imaging.