ARTIFICIAL INTELLIGENCE IN MAMMOGRAPHY: CURRENT APPLICATION AND FUTURE PERSPECTIVES

Mammography remains the cornerstone of breast cancer screening, yet its interpretation is demanding, repetitive, and susceptible to perceptual error. Artificial intelligence (AI) in mammography has progressed from retrospective algorithm development to prospective clinical evaluation and large-scale real-world implementation. Contemporary AI systems extend beyond lesion detection to support case prioritization, malignancy detection, breast-density assessment, image-quality evaluation, and risk prediction. Recent prospective and real-world studies, including MASAI, ScreenTrustCAD, PRAIM, and AI-STREAM, suggest that appropriately integrated AI can improve cancer detection and reduce reading workload without necessarily increasing false-positive outcomes. However, AI performance is context-dependent and may vary according to population characteristics, cancer prevalence, imaging systems, acquisition protocols, decision thresholds, reader behavior, and the specific role assigned to AI within the screening pathway. Important challenges remain, including generalizability, automation bias, overdiagnosis, algorithmic fairness, regulatory oversight, cybersecurity, liability, and cost-effectiveness. This narrative review critically examines current clinical applications and prospective evidence for AI in mammography, discusses key barriers to responsible implementation, and outlines priorities for future integration. Future progress should emphasize prospective multicenter validation, local performance monitoring, transparent governance, appropriate threshold selection, and demonstration of meaningful patient benefit. Rather than replacing breast radiologists, AI is more appropriately positioned as a monitored clinical decision-support tool within an increasingly personalized and efficient breast-screening pathway.