APPLICATIONS OF ARTIFICIAL INTELLIGENCE IN THE ASSESSMENT OF MAMMOGRAPHY TECHNICAL QUALITY: A SYSTEMATIC REVIEW

Abstract

The technical quality of mammography directly influences the early detection of breast cancer, with the repeat rate (RR) serving as an important indicator of quality in diagnostic imaging services. This study aimed to synthesize the scientific evidence on the application of Artificial Intelligence (AI) techniques, with an emphasis on deep learning (DL) models, in the assessment of mammography technical quality and their potential to reduce the RR. A Systematic Literature Review (SLR) was conducted following the PRISMA 2020 guidelines, with searches performed in ACM Digital Library, IEEE Xplore, PubMed, ScienceDirect, Scopus, SpringerLink, and Web of Science. The methodological quality of the studies was assessed using criteria adapted from Kitchenham. After the selection process, 13 primary studies were included in the review. DL-based models predominated, particularly Convolutional Neural Networks (CNNs) and U-Net-derived architectures, applied to anatomical landmark identification, technical image quality classification, and the implementation of decision-support and feedback systems during mammographic acquisition. The studies reported accuracy of up to 98.5% for pectoral muscle identification, a reduction in the proportion of inadequate images from 13.3% to 3.2%, and a decrease in the RR to 0.17% in the best-performing scenario identified. The evidence indicates that AI has strong potential to support mammography quality control by improving the standardization of assessments, reducing interobserver variability, and enhancing the image acquisition workflow. These findings reinforce the potential of AI as a valuable support tool for mammography quality assurance programs. However, its widespread adoption still depends on multicenter clinical validation, greater model interpretability, and the standardization of evaluation protocols.