Indiana University

Radiologists faced challenges in accurately detecting breast cancer in digital breast tomosynthesis (DBT) scans, with high false-positive rates leading to unnecessary patient callbacks. Rising breast cancer rates, particularly among younger women, increased the urgency for more accurate and efficient screening workflows.

iCAD's ProFound AI deep-learning system was implemented to support radiologists interpreting 3D mammography (DBT) scans. The system flags suspicious findings while integrating into existing radiologist workflows, enabling AI-assisted review of DBT cases across more than 16,000 exams.

Radiologists using ProFound AI identified 65% more cancers compared to without AI. Positive Predictive Value for abnormal interpretations (PPV1) doubled from 4.2% to 8.8%, and PPV3 for biopsies improved from 32% to 57%. The abnormal interpretation rate dropped from 8.2% to 6.5%, reducing unnecessary callbacks.

• Deep-learning AI integrated into radiologist workflows can simultaneously improve cancer detection rates and reduce false positives, resolving a traditional sensitivity-specificity tradeoff.
• PPV improvements (PPV1 doubling, PPV3 near doubling) indicate AI helps radiologists act on findings with greater confidence, reducing unnecessary biopsies.
• Real-world peer-reviewed studies across 16,000+ cases validate AI imaging tools as clinical assets, not just research tools.