AI-Assisted Mammography

AI-assisted mammography employs deep convolutional neural networks (CNNs) trained on large annotated datasets to identify suspicious lesions, calculate cancer risk scores and assist radiologist interpretation. Modern algorithms (Lunit INSIGHT MMG, Therapixel MammoScreen, Hologic Genius AI, ScreenPoint Transpara) achieve standalone AUC values of 0.90-0.96 comparable to expert radiologists.

Workflow integration models include: AI-as-second-reader (replacing one of two radiologists in double-reading systems prevalent in Europe), AI-as-triage (prioritizing high-risk cases for expedited review), AI-as-decision-support (highlighting suspicious findings within radiologist workflow), and AI-as-standalone-pre-screen (filtering very low-risk cases from radiologist review entirely).

Major prospective studies (MASAI Sweden, ScreenTrustCAD Sweden, PERFORMS UK, EVOLVE Korea) demonstrate AI augmentation improves cancer detection rate by 13-20% while reducing reading workload by 30-50%. False-positive recalls remain similar or slightly improved. Limitations include performance variation across breast density, manufacturer dependencies, regulatory approval scope (2D, 3D DBT, both), interpretability challenges and need for prospective real-world validation. Integration with risk models (Tyrer-Cuzick, BCSC, AI-based polygenic risk) enables personalized screening intervals.