APPLYING OF MULTIDIMENSIONAL MODELS FOR THE DIAGNOSTICS OF RETINA PATHOLOGIES

Abstract

Background. Diabetic retinopathy as a complication of diabetes mellitus and age-related macular degeneration are the most common causes of vision loss in the working population. The main features of these diseases include their asymptomatic course in the early stages of development. Early diagnosis of retinal pathologies and, as a result, early complex treatment, can greatly slow down the progression. The diagnosis is made by highly qualified specialists, but with the development of machine learning, there is the possibility of automated screening of retinal pathologies on digital images of optical coherence tomography of the eye. The purpose of this study is to develop a mathematical model for diagnosing retinal pathology in optical coherence tomography images of the eye. Materials and methods. An open database of images of optical coherence tomography of the eye was used as materials. All images were classified into four classes, of which three classes are different retinal pathologies. The following models were used as multidimensional models: logistic regression, naive bayes classifier, random forest, nearest neighbors, support vector machine, neural network, convolutional neural network. Results. Most of the models showed accuracy below 70%. The best result is obtained by the convolutional neural network with a result of 89.7%, followed by the support vector machine with an accuracy of 71.1% and the neural network with 70.8%. Conclusion. As a result, the use of multidimensional models, in particular convolutional neural networks, can show a high accuracy value, which allows using this model as a program to support medical decision making.