Using Artificial Intelligence to Predict Whether You Will Develop Breast Cancer

Radilogy published research on 17 December 2019 showing a widely discussed AI type, deep learning , predicted future breast cancer risk more accurately than density models. Dr. Karin Dembrower and colleagues at Stockholm's Karolinska Institute developed a risk score and retrospectively compared it with density-based models. Development used diagnoses from 2008 to 2012; each woman's earliest available digital mammogram was assigned a deep-learning score, dense area, and percent density.

Deep learning systems use human-made algorithms to collect, sort, and analyze large datasets. Big data feeds an artificial neural network , algorithms inspired by the brain, for interpretation. Programmers do not tell machines what to do; they analyze data and decide based on learning. More data generally improves accuracy. The study evaluated 2.283 women, with 278 later diagnosed. At mammography, future cancer patients were older (median 55,7 versus 54,6), had greater dense area (38,2 versus 34,2 cm²), and higher percent density (median % 25,6 versus 24.0 percent).

• Related topic: What are dense breasts? Why is cancer risk higher? Age-adjusted deep-learning scores were more accurate than dense area and percentage. False negatives, cancer mistakenly labeled benign, were lower, especially for aggressive cancers. Overall, the neural-network-based model outperformed density prediction. Consequently, AI decision-support systems, currently trained largely through competition with experts, are considered certain to play an important role in diagnosis and risk assessment in the near future, estimated within 10 years.