Artificial intelligence in breast cancer detection and screening: an integrative review

Abstract

Objective: To investigate advanced technologies for improving the accuracy and efficiency of breast cancer detection
and screening. Methods: An integrative review was performed on the databases PubMed, LILACS, and Embase (2001–
February 2024), using MeSH/Emtree terms: “breast neoplasms”, “artificial intelligence”, and “mass screening”. Inclusion
criteria encompassed studies on breast cancer detection/diagnosis employing artificial intelligence (AI), irrespective of
language. After duplicate removal (Rayyan software; n=31), 144 articles were screened, with 66 meeting eligibility for final
analysis. Results: AI technologies have progressed from early wavelet-based models to sophisticated deep learning systems, such as Transpara® and Lunit INSIGHT MMG, enhancing diagnostic accuracy and alleviating radiologists’ workload.
Commercial tools like Transpara® demonstrated a 15% reduction in false negatives, while Lunit INSIGHT MMG increased
early detection rates by 22%. Hybrid AI-radiologist models achieved the highest sensitivity (94%), outperforming individual
human assessments. Challenges include persistent false positives (8–12%) and scalability barriers. Emerging approaches
integrating genetic/demographic data highlight AI’s potential for personalized screening. Conclusion: AI-driven tools like
Transpara® and Lunit INSIGHT MMG demonstrate enhanced accuracy in breast cancer detection through deep learning,
yet challenges such as false positives (8–12%) and scalability persist. Personalized approaches integrating genetic/demographic data show promise for tailored screening. AI-human collaboration optimizes diagnostic accuracy; however, further validation and standardized clinical protocols are essential for widespread implementation.