Open AccessDEVELOPMENT AND VALIDATION OF OPHTHALMIC PATHOLOGY CLASSIFICATION METHOD USING DEEP MACHINE LEARNING
Author(s) -
Vadim I. Loshmanov,
Alla G. Kravets
Publication year - 2021
Publication title -
vestnik astrahanskogo gosudarstvennogo tehničeskogo universiteta. seriâ: upravlenie, vyčislitelʹnaâ tehnika i informatika
Language(s) - English
Resource type - Journals
eISSN - 2224-9761
pISSN - 2072-9502
DOI - 10.24143/2072-9502-2021-2-57-65
Subject(s) - convolutional neural network , computer science , artificial intelligence , hyperparameter , machine learning , artificial neural network , deep learning , sample (material) , chromatography , chemistry
The article highlights the problem of improving the numerous practical methods of developing the new chemical compounds that can be used in pharmacology. One of the possible ways to introduce innovative methods into conducting preclinical drug tests is the application of ac-tively developing information technologies, such as data mining with methods of deep machine learning. With a huge amount of data accumulated over several years of preclinical research, the practical solutions in this area allow to obtain a neural network data model with greater degree of accuracy, though, there is no universal method that would allow a comprehensive approach to the problem of analyzing the results of preclinical laboratory studies of drugs. Existing solutions have
a few disadvantages, which prevents from using them in practice. The two main problems are: the difficulty in verifying the results and the incompleteness of the list of calculated parameters. A system of identifying the pharmacological activity of a new drug is proposed to solve the problem, which was considered on the example of ophthalmic preclinical laboratory studies. As part of the system development, a method for classifying ophthalmic pathology based on a convolutional neural network has been implemented. The architecture of the neural network has been developed, its hyperparameters being matched experimentally. The model accuracy during training made 90%, and the test sample accuracy made 81%.