Convolutional Neural Network Analysis of Fundus for Glaucoma Diagnosis

Sabina Rakhmetulayeva, Zarina Syrymbet

2022

Closed access Metadata (CSL-JSON)
DOI 10.1109/sist54437.2022.9945723
Data confidence 70 medium
How this score is calculated

It is built from checkable properties of the record itself: whether the article has a DOI, how many independent sources confirmed its data, whether the required fields are filled in, and how recent the latest confirmation is. Disagreement between sources and a missing identifier lower it. A value of 80 or above means the record was published without further checking; 60 to 79 means it is published at reduced confidence, with a manual check advisable. A record verified by an editor by hand is never pushed below 90 by a recalculation. This scores our data about the article, not the article itself.

Abstract

Glaucoma is a disease that can progress rapidly and lead to complete loss of vision. Ophthalmologists use modern research methods to detect glaucoma in its early stages, and optical coherence tomography is currently the most accurate imaging method. With all its wide possibilities, there are cases when glaucoma can be misdiagnosed due to a limited number of parameters. The introduction of a trainable artificial intelligence network into medical diagnostics allows the use of an infinite number of variables. This technology greatly expands the possibilities of monitoring the disease and its treatment. As part of this work, we compared the criteria for determining glaucoma with each other in order to choose the one that would show a high result with an increase in data processing speed, creating a deep learning system for automatic diagnosis of glaucoma with a convolutional neural network for classification and obtaining prediction accuracy of up to 85%. Based on the results of comparing the data with each other, in order to obtain more accurate results, it was customary to use the characteristics of the fundus as a parameter for determining glaucoma. The results of the model indicated that the model we trained showed an accurate result, which was confirmed using the Custom Vision AI program, where it showed which of the images had glaucoma with an accuracy of 84% and which had a normal image with an accuracy of 94.4%.