МАММОГРАФИЯЛЫҚ СУРЕТТЕРДЕ СҮТ БЕЗІ ҚАТЕРЛІ ІСІГІН ДӘЛ АНЫҚТАУҒА АРНАЛҒАН БЕЛГІЛЕР ВЕКТОРЫН ҚҰРУ

Abstract

This paper presents a hybrid method for breast cancer classification based on Digital Imaging and Communications in Medicine (DICOM) mammography images. The method is based on the combination of deep learning techniques and textural and geometric characteristics of images. The aim of the study is to integrate textural and geometric characteristics into the neural network architecture to improve the classification accuracy and localization of abnormal areas in medical images. The relevance of the study is due to the growing incidence of cancer and the need to improve the diagnostic accuracy in resource-limited healthcare settings. The proposed method is based on a modified Faster Region-Based Convolutional Neural Network (Faster R-CNN) architecture that includes additional feature extraction units such as eccentricity, texture metrics, entropy, and mean intensity. The image preprocessing stage includes look-up table (LUT) transforms, normalization, and histogram equalization. These changes allow us to improve the key structural features of the image. The efficiency of the method was further improved by integrating texture and geometric characteristics, which reduced the overall cost by 30% and improved the quality of classification and localization. The developed method is a promising solution that will contribute to the improvement of medical diagnostics.