Generative adversarial network and texture features applied to automatic glaucoma detection

• Published in: Applied soft computing 2020-05, Vol.90, p.106165, Article 106165
• Main Authors: Bisneto, Tomaz Ribeiro Viana, de Carvalho Filho, Antonio Oseas, Magalhães, Deborah Maria Vieira
• Format: Article
• Language: English
• Subjects: Generative adversarial network; Glaucoma; Taxonomic diversity indices; Texture features
• ISSN: 1568-4946; 1872-9681
• DOI: 10.1016/j.asoc.2020.106165

Glaucoma is a neurodegenerative disease that has a multifactorial etiology. The main characteristic of this illness is the progressive lesion of the optic nerve. This disease is chronic and causes permanent blindness at an advanced stage. Early diagnosis is essential to ensure a favorable prognosis and improve the patient’s quality of life. Digital Image Processing together with computational techniques of machine learning allow the creation of methods for automatic detection of glaucoma. In this context, this work aims at the early diagnosis of glaucoma through a Generative Adversarial Network allied to texture attributes defined from indexes of taxonomic diversity. The method we propose can be divided into: (i) image acquisition through the RIM-ONE and Drishti-GS public databases; (ii) training of a conditional Generative Adversarial Network for segmentation of the optical discs into retinal images; (iii) pre-processing through enhancement and hole fill-in techniques; (iv) extraction of texture attributes using the index of taxonomic diversity; and (v) validation of the proposal through three classifiers evaluated according to four performance metrics. The results are promising and indicate that the method is robust, initially reaching 77.9% accuracy. However, as we apply improvements and adjustments in the method employed, we reach 100% accuracy and a ROC curve of 1. Therefore, we propose a second opinion on the diagnosis of glaucoma, assisting the specialist precisely.