Multiple Ocular Diseases Classification with Graph Regularized Probabilistic Multi-label Learning

Abstract

AbstractGlaucoma, Pathological Myopia (PM), and Age-related Macular Degeneration (AMD) are three leading ocular diseases in the world. In this paper, we proposed a multiple ocular diseases diagnosis approach for above three diseases, with Entropic Graph regularized Probabilistic Multi-label learning (EGPM). The proposed EGPM exploits the correlations among these three diseases, and simultaneously classifying them for a given fundus image. The EGPM scheme contains two concatenating parts: (1) efficient graph construction based on k-Nearest-Neighbor (k-NN) search; (2) entropic multi-label learning based on Kullback-Leibler divergence. In addition, to capture the characteristics of these three leading ocular diseases, we explore the extractions of various effective low-level features, including Global Features, Grid-based Features, and Bag of Visual Words. Extensive experiments are conducted to validate the proposed EGPM framework on SiMES dataset. The results of Area Under Curve (AUC) in multiple ocular diseases classification outperform the state-of-the-art algorithms.KeywordsScale Invariant Feature TransformOcular DiseaseFundus ImagePathological MyopiaColor MomentThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.