A.N. Narkevich1, T.Н. Mamedov2, D.V. Dzyuba3

1–3 FSBEI НЕ Prof. V.F. Voino-Yasenetsky Krasnoyarsk State Medical University MOH Russia (Krasnoyarsk, Russia)

Diabetic retinopathy is one of the most common complications of diabetes mellitus. This complication affects the blood vessels of the retina of the eye and can deprive the patient of vision. Due to the almost asymptomatic course, retinopathy is detected only in the late stages. In this regard, the accurate and early diagnosis of diabetic retinopathy is an urgent issue in modern medicine.

There are different methods of diagnosing diabetic retinopathy. At the moment, information technologies are developing at a rapid pace. Increasingly, complex computer technologies that are based on machine learning methods are being used in medicine. The use of neural networks is extremely promising in the screening diagnosis of diabetic retinopathy.

The ability to detect signs of diabetic retinopathy on digital images of the fundus using machine learning methods will allow automating and simplifying the diagnosis of such complications in general. There are many neural network architectures, but convolutional neural networks show the greatest efficiency for image classification.

As part of this work, the goal was to develop a model of a deep convolutional neural network for recognizing diabetic retinopathy on digital images of the fundus.

The work uses a set of fundus images from the open database of the Kaggle website. In total, the set is represented by 12,498 images. Among them, there were 7,914 images showing the retina of the eye with signs of DR and 4,584 images without signs of DR. After preprocessing for optimization and more efficient training of the neural network, the initial number of images was increased by 360 times.

Training of the convolutional neural network model made it possible to achieve a high accuracy of classification of diabetic retinopathy on digital images of the fundus, which was 93.4 [92.6; 94.2]%. This allows us to use this model for screening diagnostics of diabetic retinopathy in patients with diabetes mellitus.

Наркевич А.Н., Мамедов Т.Х., Дзюба Д.В. Распознавание диабетической ретинопатии на цифровых изображениях глазного дна с применением сверточных нейронных сетей глубокого обучения // Технологии живых систем. 2023. T. 20. № 1. С. 55-61. DOI: https://doi.org/10.18127/j20700997-202301-06

Narkevich A.N., Mamedov T.H., Dzyuba D.V. Recognition of diabetic retinopathy on digital images of the fundus using convolutional neural networks of deep learning. Technologies of Living Systems. 2023. V. 20. № 1. Р. 55-61. DOI: https://doi.org/10.18127/j20700997-202301-06 (In Russian)