APPLICATION OF NEURAL NETWORK TECHNOLOGIES FOR ENHANCING THE INFORMATIVENESS OF MEDICAL IMAGES

Abstract

The paper considers modern approaches to the use of artificial neural networks for improving the quality and informativeness of medical images in intrascopic diagnostic systems. In particular, a method and tool are presented for enhancing the quality of dental intrascopic images based on the integration of classical digital processing algorithms and deep neural networks. Special attention is paid to the analysis of convolutional neural networks (CNN), the U-Net architecture, and generative adversarial networks (GAN). An overview of the advantages and limitations of their application in medical tasks is provided. The relevance of the study is determined by the need for more accurate visualization of fine dental and oral tissue structures, which is critically important for the early diagnosis of dental diseases. The proposed approach consists of three key stages: preliminary image preprocessing (noise reduction and brightness normalization), data reconstruction using the U-Net architecture, and evaluation of the results through objective metrics (PSNR, SSIM) as well as subjective expert assessments by dentists. The results of prototype modeling of the neural network method aimed at noise reduction and image contrast enhancement are presented. Experimental studies demonstrated that the proposed method increases PSNR by 15–20 % and improves SSIM by an average of 0.1–0.15 compared to classical filtering methods. The obtained results confirm the effectiveness of the combined approach for processing dental intrascopic images. The method allows preserving fine anatomical details, reducing the influence of noise and artifacts, and increasing image informativeness, which in turn creates prerequisites for more accurate clinical diagnosis and decision support in dental practice.