Sposob generatisi syntheticskih medicinskih isobrajeni s primeneniem modulnoi generativno-sostyazatelnoi architektury

Abstract

The invention relates to the field of information technologies, in particular to methods of image generation using artificial intelligence in medical visualization. A method of generating synthetic medical images using a modular generative adversarial architecture is proposed, consisting of interconnected functional modules: a data preprocessing module, a generator with a cascade of transforming layers, a discriminator with a built-in diagnostic submodule based on a pre-trained neural network VGG19, a learning module with the implementation of the WGAN-GP strategy, and a quality assessment module using metrics KID, FID, LPIPS and MS-SSIM. The method is aimed at generating images of small (minority) pathologies for the purpose of balancing training samples, improving the quality and stability of models of automated diagnostics, as well as expanding the possibilities of analysis in clinical practice.