"A Transformer-Based Deep Learning Framework for Multimodal Glaucoma Management"

"Glaucoma is a progressive eye disease that can cause irreversible vision loss if not managed effectively. Predicting disease progression is crucial for timely intervention. This study aims to design a Transformer-based two-stage pretraining model to enhance the prediction of visual field (VF) maps from color fundus images, addressing key challenges in glaucoma management. To overcome data scarcity, we curated the Gla-2K dataset for glaucoma pretraining. The model uses a two-stage pretraining framework: the first stage focuses on learning general ophthalmic features from a large set of data, while the second stage targets glaucoma-specific features. Transformer-based deep learning models integrate multimodal data, including age, intraocular pressure, central corneal thickness, and other ophthalmic factors. The method significantly improved VF progression prediction. Compared to the baseline models, it achieved a 0.568 reduction in RMSE, a 0.592 decrease in MAE, and a 0.121 increase in R\u00b2, demonstrating superior performance. Clinical validation confirmed the model's practical applicability. The two-stage pretraining framework provides an effective approach for glaucoma management. By integrating multimodal data and leveraging pretraining techniques, the model establishes a new benchmark, offering substantial potential for improving glaucoma diagnosis and long-term management."