Evaluation of machine learning classifiers in keratoconus detection from orbscan II examinations

PURPOSE: To evaluate the performance of support vector machine, multi-layer perceptron and radial basis function neural network as auxiliary tools to identify keratoconus from Orbscan II maps. METHODS: A total of 318 maps were selected and classified into four categories: normal (n = 172), astigmatism (n = 89), keratoconus (n = 46) and photorefractive keratectomy (n = 11). For each map, 11 attributes were obtained or calculated from data provided by the Orbscan II. Ten-fold cross-validation was used to train and test the classifiers. Besides accuracy, sensitivity and specificity, receiver operating characteristic (ROC) curves for each classifier were generated, and the areas under the curves were calculated. RESULTS: The three selected classifiers provided a good performance, and there were no differences between their performances. The area under the ROC curve of the support vector machine, multi-layer perceptron and radial basis function neural network were significantly larger than those for all individual Orbscan II attributes evaluated (p<0.05). CONCLUSION: Overall, the results suggest that using a support vector machine, multi-layer perceptron classifiers and radial basis function neural network, these classifiers, trained on Orbscan II data, could represent useful techniques for keratoconus detection.

Diagnóstico do ceratocone baseado no Orbscan com o auxílio de uma rede neural / Detection of keratoconus based on a neural network with Orbscan

OBJETIVO: Desenvolver uma rede neural artificial para classificar em normal ou portador de ceratocone os pacientes submetidos ao exame do Orbscan II TM. MÉTODOS: Foi realizado um estudo retrospectivo envolvendo 98 exames de 59 pacientes. Utilizando o programa Java Neural Network 1.1 foi criada uma rede neural artificial para classificar os exames entre os dois grupos (normais e portadores de ceratocone). Foram utilizados 73 exames para treinamento e validação da rede, e 25 para testar o seu funcionamento. RESULTADOS: Dos 73 exames utilizados no treinamento da rede, 59 eram normais e 14 mostravam alterações relacionadas ao ceratocone. O método utilizado para treinamento da rede foi o "backpropagation". A taxa de aprendizado utilizada foi de 0,2, e a taxa de tolerância de erro 0,05. Dos 25 exames utilizados para a avaliação da eficácia da rede, 19 eram normais, e 6 apresentavam ceratocone. Após o treinamento a rede apresentou sensibilidade e especificidade de 83 e 100 por cento, respectivamente. CONCLUSÃO: A rede neural artificial representa uma opção útil e viável para auxiliar na classificação de exames realizados com o Orbscan II TM.

PURPOSE: To evaluate an artificial neural network in order to correctly identify Orbscan II TM tests of patients with normal and keratoconus corneas. METHODS: A retrospective analysis included 98 Orbscan II TM tests of 59 subjects and an artificial neural network was created and trained based on the Java Neural Network 1.1 software. Seventy-three tests (59 normal tests and 14 keratoconus examinations) were applied to train the neural network and 25 eyes were used to test the method (19 normal eyes and 6 cases of keratoconus corneas). RESULTS: Backpropagation method was performed to train the neural network to 5 percent error and 0.2 learning rate. The trained neural network presented sensibility and specificity of 83 and 100 percent respectively. CONCLUSION: Artificial neural network can accurately help clinicians to classify keratoconus in Orbscan II TM tests.