Nearly half of women have dense breasts, a risk factor for breast cancer. For proper screening, spotting dense breast tissue is important, as it can hide the presence of tumors. Currently, dense tissue is identified by radiologists viewing mammography images, but their evaluations are subjective and therefore can vary from physician to physician.
A team of scientists from MIT and Massachusetts General Hospital have now created a piece of software that can automatically assess whether a given woman’s breast tissue is dense or not, with an accuracy matching that of a team of human radiologists.
“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,” said Adam Yala, one of the leads of the research.
The system relies on a convolutional neural network, a computer science technique now widely used in a number of fields including image recognition tasks. It was trained, using nearly 60,000 mammography images, to learn what dense breasts look like thanks to each image having been already been used in a real case and tagged with a radiologist’s diagnosis.
The system was tested on patients screened at the Massachusetts General Hospital, with radiologists receiving the software’s estimate of breast density with every mammography image. The radiologists were simply given a choice to agree or to disagree with the system’s decision, and after looking at over 10,000 images the system demonstrated a 94% agreement with the radiologists.
Study in journal Radiology: Mammographic Breast Density Assessment Using Deep Learning: Clinical Implementation…
Via: MIT…