Diagnostic Accuracy of Artificial Intelligence for Breast Cancer Detection: A Systematic Review

Abstract

Background: Breast cancer remains a leading cause of cancer-related morbidity and mortality worldwide, necessitating accurate and early detection strategies. Conventional imaging and pathological assessment are limited by interobserver variability, reduced sensitivity in dense breast tissue, and increasing workload pressures. Artificial intelligence (AI) and machine learning (ML) have emerged as potential tools to enhance diagnostic performance and clinical decision-making. Objective: To systematically evaluate the diagnostic accuracy and clinical applicability of Artificial Intelligence & Machine Learning models in breast cancer detection across imaging modalities and clinical settings. Methods: A systematic search of PubMed, Scopus, and Web of Science was conducted for studies published between January 2014 and May 2025, following PRISMA 2020 guidelines. Original studies assessing AI/ML-based diagnostic models and reporting performance metrics were included. Two reviewers independently performed study selection, data extraction, and risk-of-bias assessment using the Prediction Model Risk of Bias Assessment Tool (PROBAST). Due to substantial methodological heterogeneity, results were synthesized descriptively. Results: Of 1,892 identified records, 29 studies met inclusion criteria. Most evaluated imaging-based models using mammography, ultrasound, MRI, CT, thermography, or digital histopathology. Deep learning approaches, particularly convolutional neural networks, predominated. Reported AUC values ranged from 84 to 99%, with sensitivity and specificity frequently exceeding 85% in retrospective cohorts. Large screening studies demonstrated that AI-assisted mammography was non-inferior to double reading while reducing radiologist workload. However, most studies relied on retrospective datasets with limited external validation. Conclusion: Artificial Intelligence & Machine Learning models show high diagnostic potential across breast imaging modalities and may enhance screening efficiency and diagnostic support. Nevertheless, the predominance of retrospective designs and limited prospective multicentre validation restricts assessment of real-world generalizability. Rigorous external validation, standardized reporting, and implementation-focused research are essential before widespread clinical integration.