Glaucoma is an asymptomatic ocular disease at the onset which, if left untreated, can lead to blindness. The World Health Organization (WHO) has estimated that by 2020 glaucoma should affect 80 million people and by 2040 will be 111.5 million. In this context, the present work aims to compare methods of automatic classification of glaucoma to aid in its diagnosis by medical specialists. For this, two models are developed based on the extraction of characteristics of eye fundus images obtained from the public base RIM-ONE v2, and also uses a simple convolutional neural network (CNN). In the first model, texture characteristics, histogram information and image colors are extracted, which are submitted to the Principal Component Analysis (PCA) to evaluate nine classical classifiers including Multilayers perceptron (MLP). In the evaluation are used Accuracy, sensitivity, specificity and area under curve (AUC). The results demonstrate that the best performance is obtained with the MLP classifier with PCA use in the characteristic descriptors, with accuracy of 90.11%, sensitivity 97.44%, specificity of 84.62%, and AUC equal to 0.90. Resumo: O glaucomaé uma doença ocular assintomática no início que, se não for tratada a tempo, pode levar a cegueira. A Organização Mundial de Saúde (OMS) estimou que em 2020 o glaucoma deve afetar 80 milhões de pessoas e em 2040 serão 111,5 milhões. Neste contexto, o presente trabalho visa comparar métodos de classificação automática do glaucoma para auxiliar no seu diagnóstico por médicos especialistas. Para tanto, são desenvolvidos dois modelos com base na extração de características de imagens de fundo de olho, obtidas da base pública RIM-ONE v2, e também se utiliza de uma rede neural convolucional (CNN) simples. No primeiro modelo, extraem-se características de textura, informações de histogramas e cores das imagens, que são submetidasà Análise de Componentes Principais (PCA) para, em seguida, realizar a avaliação de nove classificadores clássicos, incluindo o Multilayers perceptron (MLP). No segundo modelo, adicionam-se características oculares geométricas extraídas: Cup to Disk Ratio (CDR), Neuro-Retinal Rim (NRR) e Blood Vessel Ratio (BVR). Empregam-se na avaliação a acurácia, sensibilidade, especificidade eárea sob a curva (area under curve-AUC). Os resultados demonstram que o melhor desempenhoé obtido com o classificador MLP com uso do PCA nos descritores de características, com acurácia de 90,11%, sensibilidade 97,44%, especificidade de 84,62%, e AUC igual a 0,90.

Glaucoma is an asymptomatic eye disease that, if not treated, can lead to blindness. The World Health Organization (WHO) estimated that by 2020 glaucoma should affect 80 million people and by 2040 will be 111.5 million. In this context, the present work aims to compare automatic classification methods to assist the specialist physician in the diagnosis of glaucoma. For this purpose, a model based on the extraction of nongeometric characteristics of optical disk images from the RIM-ONE r2 dataset had been developed. These characteristics were submitted to Principal Component Analysis (PCA) for dimensionality reduction, the resulting components served as input to the classifiers: Logistic Regression (RL), Decision Tree Gradient Boosting (DTGB), Support Vector Machine ( SVM), k nearest neighbors (k-NN) and Multilayer Perceptron (MLP). To evaluate the results we used the accuracy, sensitivity, specificity, positive predictive value, negative predictive value and area under curve (AUC). The results demonstrate good performance with all classifiers, especially MLP, the test results reached an accuracy of 97.83 %, sensitivity 100 %, specificity 96.15 %, vpp 95.24 %, vpn 100 % and auc 97.62 %.