Abstract:
|
Longitudinal fundus images provide valuable insights into the progression of age-related macular degeneration (AMD), which is a progressive eye disease and a leading cause of vision loss in the developed world. However, the existing prediction models for AMD progression have not fully incorporated longitudinal fundus photographs. In this research, we develop a deep-learning-based survival model for the dynamic prediction of AMD progression trajectories using longitudinal fundus images. Specifically, the trajectory of the prognostic index for each patient is fitted via a RNN with longitudinal images as input. Then we obtain the individualized prognostic index trajectory and use that to predict disease progression risk. We establish and evaluate our models on 52,610 fundus images of 4,318 participants from the Age-Related Eye Disease Study (AREDS), in which fundus images and disease severity scores are captured at baseline and follow-up visits over a period of 12 years. The prediction performance of each model is evaluated on two accuracy metrics: time-dependent Brier Score and C-index. The results indicate that our model achieves satisfactory prediction performance.
|