Researchers at MIT University have created a deep learning model that can predict the risk of breast cancer in women up to five years in advance. To do this, the model looks at subtle tissue patterns in breasts that occur in the early phase of the disease.
To train the model, researchers used the mammograms of more than 30,000 Massachusetts General Hospital patients whose medical consequences were known. As a result, the model learned to recognize subtle tissue patterns that are invisible to the human eye and may indicate future cancer cells. By looking at those patterns in patients, the model can indicate whether a patient has an increased risk of breast cancer in the period of three to five years after the mammography.
The deep learning model is more accurate than the Tyrer-Cuzick model, which is now used by medical professionals to predict breast cancer risk in women, the researchers say. This TC model looks at the breast tissue density and life characteristics of the woman. Of the 60,000 patients, 269 are known to have or have had breast cancer. Of those 269, the new MIT model placed 31 percent in a high-risk group. The TC model would have placed only 18.2 percent of those women in a high-risk group.
An additional advantage of the model is that it works better for minorities. The existing models are based on characteristics such as age, hormonal factors, family histories and breast tissue density. These models are often made on the basis of data from white women. As a result, they are less accurate for black women, for example. In America, this target group is 43 percent more likely to die from breast cancer than white women. Because the MIT model is based on the tissue patterns of more than 30,000 patients, a woman’s skin color is irrelevant.
Based on the findings of this model, personal screening appointments could be made with patients. For example, the doctor may recommend additional tests for a woman who, according to the model, has an increased risk of breast cancer. According to the researchers, there have been doctors who have argued for individual screening appointments for some time, but this would not be possible because tests are not accurate enough for this. This model would be accurate enough for that, the researchers said.
The research is published under the title A Deep Learning Mammography-based Model for Improved Breast Cancer Risk Prediction.