Multi-Stage Ensemble-Based System for Glaucomatous Optic Neuropathy Diagnosis in Fundus Images

Abstract

Recent developments in Computer-aided Diagnosis (CAD) systems as a countermeasure to the increasing number of untreated cases of eye diseases related to visual impairment (such as diabetic retinopathy or age-related macular degeneration) have the potential to yield in low-to-mid income countries a comfortable and accessible alternative to obtaining a general ophthalmological study necessary for follow-up medical attention. In this work, a multi-stage ensemble-based system for the diagnosis of glaucomatous optic neuropathy (GON) is proposed. GON diagnosis is based on a binary classification procedure working in conjunction with a multi-stage block based on image preprocessing and feature extraction. Our preliminary data show similar results compared to current studies considering metrics such as Accuracy, Sensitivity, Specificity, AUC (AUROC), F1score, and the use of Matthews Correlation Coefficient (MCC) as an additional performance metric is proposed.