Summary: We propose a method that incorporates both state-of-the-art deep learning-based image extraction features as well as standard metrics used in clinical practice to provide accurate Glaucoma detection irrespective of imaging and clinical conditions. Embeddings extracted by a ResNet-50 model on a fundus image as well as numerical features indicative of Glaucoma (CDR, cup eccentricity, and disk eccentricity) are concatenated to provide a comprehensive diagnosis.
Authors: Sauman Das, Arnav Jain, Audhav Durai, Sameer Gabbita, Aditya Vasantharao, Vishal Kotha