Prediction of Cup-to-Disc Ratio Progression Based on Longitudinal Datasets of Glaucoma Patients

Glaucoma is the first irreversible cause of blindness. People have been working hard on the prediction of glaucoma and the progression of visual field, the segmentation of optic cup and disc, and the influencing factors of the change of cup-disc ratio, but there are few studies on the prediction of cup-disc ratio. In this paper, we propose an artificial intelligence model to predict future progress in the cup-to-plate ratio of glaucoma patients based on individual basic data and longitudinal data related to their disease. The main challenge is that the existing data sets have problems with time alignment. To solve this challenge, we introduce a method called feature transformation. In addition, we also use data smoothing methods to deal with data noise and improve model generalization. The experimental results show that our model has ideal effect and excellent generalization. After data smoothing, in test sets, R₂ of cross-validation (k=10) is from 0.97 to 0.98, MSE of cross-validation (k=10) is from 0.0003 to 0.0005. In the external validation sets, R₂ is 0.85 and the MSE is 0.0034.