According to a study published in the journal Radiology, a complex type of artificial intelligence (AI) can outperform existing models in predicting the risk of developing cancer of the breast.
Most existing breast cancer screening programs based on mammography at regular intervals to all women. This approach is not optimized for the detection of cancer at the individual level and can reduce the effectiveness of screening programs.
"Risk Forecasting is an essential component of individual screening policy - said study lead author Karin Dembrover, MD, a radiologist at the Karolinska Institute in Stockholm, Sweden -. Effective risk forecasting can improve confidence in screening programs."
High breast density or more glandular and connective tissue compared to fat is considered to be risk factor for cancer. Although the density may be included in the risk assessment, the existing forecasting models can not make full use of the information contained in mammography. This information can identify women who will benefit an additional screening with MRI.
Dr. Dembrover and his colleagues have developed a risk model based on a deep neural network, the type of AI that can extract a lot of information from the mammography images. He has significant advantages compared to methods such as visual assessment of breast density radiologist, who can not always take into account all relevant risk information in the image.
The new model is based on cases diagnosed between 2008 and 2012, and then studied in more than 2,000 women aged 40 to 74 years who underwent mammography at the Karolinska University. Of the 2283 women who participated in the study, 278 were later diagnosed with breast cancer.
The underlying neural network showed higher accuracy in breast cancer risk compared with mammography. False-negative rate in the neural network at the same time was lower.
"The underlying neural networks generally provide better performance than the model based on the density, - says Dr. Dembrover -. A more aggressive subtypes of cancer do not have a negative impact on its predictive accuracy."
Results of the study confirm the future role of AI in the assessment of breast cancer risk. "We do not report the mammographic density, - says Dr. Dembrover -. At individual screening we use deep learning networks that can predict cancer, but do not choose an indirect way, which offer densities."
As an added benefit, the AI approach can constantly improve by providing more high-quality data sets.
"Our experts on deep training at the Royal Institute of Technology in Stockholm, working to improve the model, - says Dr. Dembrover -. After that, we plan clinical testing models, offering MRI for women who get the most benefit from it."