Breast density assessments have traditionally relied on subjective human exams and calculations, but the deep-learning model -- trained on tens of thousands of digital mammograms -- is able to distinguish different types of breast tissue, from fatty to extremely dense, with 90 percent correlation to radiologists' diagnosis.
In comparison to traditional prediction models, the researchers used a metric called a kappa score, where 1 indicates the model and human experts agree on a diagnosis every time, and anything lower indicates fewer instances of agreements. The maximum kappa score for existing automatic density assessment models is around 0.6. In the clinical application, the new model scored 0.85, meaning it makes better predictions than previous systems.
"Breast density is an independent risk factor that drives how we communicate with women about their cancer risk. Our motivation was to create an accurate and consistent tool, that can be shared and used across health care systems," says MIT PhD student and second author on the model's paper, Adam Yala. "It takes less than a second per image ... [and can be] easily and cheaply scaled throughout hospitals." The researchers now plan on exploring how the algorithm can be transitioned into other hospitals, and how it can be used in other healthcare applications.