INFORMATION TECHNOLOGY FOR GLAUCOMA DIAGNOSIS AND PROGNOSIS

Abstract

The object of this study is the processes of automated diagnosis and prediction of glaucoma, based on the intellectual analysis of retinal fundus images and the results of optical coherence tomography. The relevance of the work is determined by the critical increase in the number of cases of blindness due to the latent course of glaucoma, which requires the introduction of effective means of early detection of the disease. The analysis demonstrated that existing diagnostic methods often have a narrow theoretical focus, high cost and require the mandatory participation of an experienced ophthalmologist to interpret complex neural network data. The purpose of the study is to develop low-resource information technology that provides high accessibility, cost-effectiveness and automation of decoding results for mass screening of the population. The work uses machine and deep learning methods, in particular the VGG-16, ResNet-50 architectures and transformer models, which allow capturing small spatial dependencies for accurate segmentation of the optic nerve. The proposed technology automates the full data processing cycle, including preprocessing, vascular mesh elimination, and binary classification of the eye condition into the categories of “normal” or “pathology”. The scientific novelty of the work lies in improving the method of reconstructing damaged areas of the retina through the design of a complex loss function that combines the parameters of contextual attention, edges, and preservation of blood vessel features. The use of a modified masked Gaussian blur method and transitive learning allowed us to significantly improve the quality of retinal image processing. The practical significance of the results obtained lies in the possibility of deploying the developed models on peripheral devices with limited resources, which allows for rapid diagnostics without the involvement of expensive equipment and minimizes the influence of the human factor on medical decision-making.