Medical News Observer

• The AI model surpasses clinical risk models in detecting high-risk breast cancer cases by analyzing mammographic features for precise predictions.
• A study across data from Italy, Spain, and Germany demonstrates the AI model’s effectiveness, identifying 6.2% of women at high risk, consistent across varying mammographic densities.
• Proven in Sweden and the USA, the AI model now shows generalizability in Europe, predicting 30% of significant breast cancers in high-risk women previously missed in screenings.

An international study led by Karolinska Institutet in Sweden, published in The Lancet Regional Health – Europe, demonstrates the effectiveness of an AI-based risk model in identifying women at high risk of breast cancer during mammography screenings. Their study evaluated the performance of an AI-based risk model for breast cancer in European screening settings, comparing it with clinical risk models that utilize family history and lifestyle factors. The model uses AI to detect minute changes in mammographic images invisible to the human eye.

Conducted across Italy, Spain, and Germany, the study involved 739 women with breast cancer and 7812 controls. The AI model analyzed mammographic features from digital mammograms to predict two-year absolute risks of breast cancer. Results showed that the AI model demonstrated good predictive accuracy. Notably, the model effectively identified 6.2% of women at high risk (as per the NICE guidelines) who were more likely to develop cancer within two years. This performance was consistent across different mammographic densities.

Previously evaluated in Sweden and the USA, the AI model has shown its generalizability across European mammography programs. It successfully predicted about 30% of clinically significant stage 2 and higher breast cancers in high-risk women initially sent home with negative mammograms. The next step involves a clinical study in Europe to assess the implementation of this model in real-world screening settings. This proposed model has the potential to enhance accuracy and decrease the workload of screening personnel.

Ref:Eriksson, Mikael, Marta Román, Axel Gräwingholt, Xavier Castells, Andrea Nitrosi, Pierpaolo Pattacini, Sylvia Heywang-Köbrunner, and Paolo G. Rossi. 2023. “European Validation of an Image-Derived AI-Based Short-Term Risk Model for Individualized Breast Cancer Screening—a Nested Case-Control Study.” The Lancet Regional Health – Europe 0 (0). https://doi.org/10.1016/j.lanepe.2023.100798.