A deep learning model, trained on thousands of mammograms, can now predict whether a person is at a high risk of developing breast cancer. Better than any current model, the new system can identify subtle changes in breast tissue and determine how likely it is that cancer could develop within five years.
The new deep learning model was trained on over 90,000 mammograms. The system was able to identify tiny patterns in the mammogram data that humans could not pick up on. The result is a current ability to identify 31 percent of patients at the highest risk of developing breast cancer in the near future. While this rate may sound low it is actually significantly better than any of the models currently available to doctors, which can only identify 18 percent of high-risk patients at an early stage.
"Since the 1960s, radiologists have noticed women have unique and widely variable patterns of breast tissue visible on the mammogram," explains Constance Lehman, co-author on the latest research. "These patterns can represent the influence of genetics, hormones, pregnancy, lactation, diet, weight loss and weight gain. We can now leverage this detailed information to be more precise in our risk assessment at the individual woman level."
Another strength of the new AI model is its reported accuracy in detecting breast cancer risk for racial minorities. Prior breast cancer risk assessment tools used by clinicians have proved less effective for non-white populations due to the fact much of the data used in development came from white populations.
"It's particularly striking that the model performs equally as well for black and white people, which has not been the case with prior risk assessment tools," says Allison Kurian, from the Stanford University School of Medicine. "If validated and made available for widespread use, this could really improve on our current strategies to estimate risk."
AI-driven diagnostic research is booming at the moment, with models being developed to identify everything from Alzheimer's and skin cancer, to child depression through speech tracking. This new MIT model adds to earlier work from the same team developing a system can predict the likelihood of a breast lesion becoming either malignant or benign with incredible accuracy.
The researchers working on the new breast cancer AI model suggest further study is needed to broadly validate the system before it is widely implemented in clinics. However, the idea that deep learning models can identify patterns in mammogram data that humans cannot seems incredibly promising. And, these models will only get more accurate as they are trained on larger sets of data, so the future looks bright for catching cancer before it grows and spreads.
The new research was published in the journal Radiology.
Source: MIT Computer Science and Artificial Intelligence Laboratory
