Image-derived artificial intelligence (AI)--based risk models for breast cancer have shown high discriminatory performances compared with clinical risk models based on family history and lifestyle factors. However, little is known about their generalizability across European screening settings. European researchers have, therefore, investigated the discriminatory performances of an AI-based risk model in European screening settings.

Using four European screening populations in three countries (Italy, Spain, Germany) screened between 2009 and 2020 for women aged 45–69, the researchers performed a nested case-control study to assess the predictive performance of an AI-based risk model. In total, 739 women with incident breast cancers were included together with 7812 controls matched on year of study-entry. Mammographic features (density, microcalcifications, masses, left-right breast asymmetries of these features) were extracted using AI from negative digital mammograms at the study entry. Two-year absolute risks of breast cancer were predicted and assessed after two years of follow-up. Adjusted risk stratification performance metrics were reported per clinical guidelines.

The research

In the trial, all women had digital mammography at the study baseline. At the next scheduled screen, where cancer outcome was assessed, half of the women were randomized into having digital mammography, arm 1, and the remaining half also had digital breast tomosynthesis (DBT), arm 2. This was done as part of the original randomized controlled trial (RCT) with the same name.For their current study, they had a different aim, where the researchers studied risk assessment of breast cancer based on the prior mammograms and reported the risk of breast cancer at the study baseline for the two outcome groups, arms 1 and 2. The Paderborn study population had no access to interval cancers.

The article followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines for cohort studies. 

Breast cancer occurrence and tumor stage were retrieved from medical records using hospital-specific personal identification numbers for each population. The American Joint Committee on Cancer (AJCC) classification defined the tumor stage.

Finding the results

In an external validation in four European screening populations, the researcher investigated the discriminatory performance and risk classification of an image-derived AI-based risk short-term model designed to identify women who are at high risk of breast cancer before or at the next screen after a negative screen. The AI-based risk model showed discrimination similar to that of the original report. Similar risk stratification performances were observed in women with dense and non-dense breasts. Late-stage breast cancers were more likely to be diagnosed in women at high risk than women at general or moderate risk.

The image-derived AI-based risk model showed a generalized performance for identifying and classifying the risk of breast cancer in four European screening populations. The model predicts clinically relevant stage 2 and higher breast cancers in women who are at high risk of breast cancer before or at the next screen and are sent home with a negative mammogram. An image-derived AI model is feasible for personalized breast cancer screening to improve the screening outcomes.