Artificial intelligence for detection of age-related macular degeneration based on fundus images: A systematic review

Age-related macular degeneration (AMD) is one of the most common types of eye diseases that generally affect the elderly population over 50 years of age. The effects of AMD on the quality of vision and life are devastating. We systematically review applications and performance of machine/deep learning algorithms for AMD detection and prediction using color fundus photos. We reviewed the studies that focused on machine learning and deep learning techniques and algorithms to analyze the fundus images for AMD. The data were collected by searching Scopus, PubMed (Medline), Web of Science and IEEE Xplore databases. After screening, 42 papers were included. The findings showed that the studies used different architectures for model training and testing. Convolutional neural networks (CNN) are used mostly in the diagnosis of AMD. ResNet architecture (11 studies) was used more than other architectures. Twenty-two studies used AREDS dataset. CNN algorithm with ResNet architecture had the highest performance compared to other architectures. Studies have shown that machine learning can diagnose AMD from fundus images with high accuracy; however, calibration, fairness, explainability, external validation, generalization, prospective validation in clinical settings and regulatory requirements should be considered in the future.