Result: Software tool for synthesis and classification of biomedical images

The paper presents a software tool for classification and synthesis of biomedical images that has been developed in the course of research. The need for artificial expansion of data sets of biomedical images is established due to their limited availability, which creates an obstacle to the development of diagnostic tools. It was found that the developed software tool can solve this problem by generating synthetic, but realistic medical images, which can serve as additional data for the training of classifiers. Functional requirements for the software tool and its architecture were developed using modern programming and software design technologies. The software tool is designed using a modular architecture that allows each module to be scaled independently according to the load. The regularities of the software architecture are characterized, including client-server interaction, the MongoDB database and the use of the RabbitMQ message broker for asynchronous data exchange between the software modules. The main modules of the software tool are as follows: datasets (responsible for managing training images), classifiers (responsible for training and using convolutional neural networks for image classification), and generators (responsible for training and using generative adversarial networks for image synthesis). The software tool was developed using various programming languages such as Python, TypeScript and modern technologies, in particular, NodeJS, RabbitMQ, PyTorch, MongoDB, React. The database structure is also designed using a logical model based on a UML class diagram. The effectiveness of using convolutional neural networks and generative-adversarial networks for classification and synthesis of biomedical images, respectively, is shown. The conclusion is made about the scientific novelty and practical significance of the developed software tool, which represents new opportunities for medical diagnostics and research, providing flexibility and scalability in