MTUSI: Deep Learning Method for Segmentation of Eye Capillaries

The main articles are:

• Sight
• Big Data in Medicine
• Artificial Intelligence in Medicine
• Neural networks (neural networks)
• Machine Learning

2023: Development of a deep learning method for segmentation of the capillaries of the eye

Scientists from MTUSI have developed a deep learning method that allows you to segment the vessels of the eye. This method will greatly simplify the work of specialists in the field of diagnosing diseases in the early stages of development, since the blood vessels of the retina are associated with many diseases: diabetes mellitus, blood clots (vascular occlusion), hypertension, stroke and others. The university announced this on November 15, 2023.

To use machine learning methods and deep learning methods, the U-Net neural network model was selected, capable of capturing the maximum number of patterns on large data arrays, which, in turn, can be quickly combined with the available data for the necessary task. U-Net shows good quality due to the implementation of skipconnection, which allows you to store some of the spatial information when the encoder compresses the image. All this contributes to better results at the lowest time and resource costs.

Of course, the deep learning method with the U-Net neural network approach to solve the problem of segmentation of the capillaries of the eye shows excellent results with a low value of the loss function. This method can be used as an intelligent assistant to ophthalmologists and various other medical professionals for further classification of diseases, - said Irina Sineva, scientific director of the development, Ph.D., associate professor of the Department of Probability Theory and Applied Mathematics at MTUSI.

According to the model developer, 1st year master's student of MTUSI Daniil Matveev, work is underway to increase the indicators of deep learning methods with the involvement of additional specialists in order to get rid of uncertainty data and clarify the results. Ideally, it will be possible to significantly improve the quality of eye capillary detection, thereby simplifying the work of specialists.