INTELLIGENT SYSTEMS AND TECHNOLOGIES

Enhancing the accuracy of dental diagnostics from X-ray images based on the neural network software system

M. D. Borodin, M. L. Rysin

MIREA — Russian Technological University, Moscow, Russia

Abstract: The article presents important aspects of the design and results of the evaluation of an intelligent decision support system for diagnostics in dental practice. A modified U-Net architecture is proposed, with a loss function optimized for detecting various pathologies in dental radiographs. The developed mathematical model for the neural network includes components of sensitivity, specificity, F-measure, binary cross-entropy, and probability calibration with adaptive weighting coefficients. The system integrates attention blocks, deep supervision, and spatial dropout, ensuring effective diagnosis of both common and rare dental pathologies. Experimental evaluation demonstrates the superiority of the developed system over existing Russian analogs based on key metrics. The implemented architecture and methods significantly enhance the reliability of predictions and, consequently, the accuracy of automated diagnostic results.

Keywords: intelligent decision support system, dental diagnostics, U-Net, neural networks, rare pathologies, probability calibration, radiographic diagnostics, automated diagnostics, machine learning.

DOI: 10.14357/20718632250303

Bibliographic databases:

•