Deep learning vs tradtional breast cancer risk models to support risk-based mammography screening

Deep learing (DL) breast cancer risk models demonstrate improved accuracy compared to traditional risk models but have not been prospectively screened. A prospective study of consecutive patients screened at five facilities in the USA compared DL risk score derived from the patient’s prior mammogram with traditional risk scores to prospectively identify patients with cancer. Cancer detected per thousand patients screened were higher in patients at increased risk by the DL model (8.6; 95% CI 7.9-9.4) compared to the Tyrer-Cuzick model (4.4; 3.9-4.9) and the CI BCRAT model (3.8; 3.3-4.3). The AUC of the DL model (0.68; 95% CI 0.66-0.70) was higher compared to Tyrer-Cuzick (0.57; 0.54-0.60) and NCI BCRAT (0.57; 0.54-0.60). The authors conclude that a CL model to assess breast cancer risk can support feasible and effective risk-based screening and is superior to traditional models to identify patients destined to develop cancer in large screening cohorts.