Deep learning model as accurate as radiologists in identifying breast density

Massachusetts General Hospital and MIT have developed a deep learning algorithm that assesses breast tissue density in mammograms.

Massachusetts General Hospital and MIT have developed a deep learning algorithm that assesses breast tissue density in mammograms.

The artificial intelligence approach is as accurate as radiologists in assessing tissue density, an independent risk factor for breast cancer, researchers contend.

A convolutional neural network was trained on tens of thousands of high-quality digital mammograms to assess four Breast Imaging Reporting and Data System (BI-RADS) density categories—fatty, scattered (scattered density), heterogeneous (mostly dense), and dense.

Researchers contend it is the first time a deep learning model of its kind has been successfully used in a clinic on real patients and has been shown to operate at the same level of experienced mammographers.

Also See: Algorithm as accurate as radiologists in assessing breast density, cancer risk

“A deep learning algorithm was used to reliably and accurately assess mammographic breast density in a large clinical practice,” conclude researchers in a study published earlier this month in the journal Radiology. “Given the high level of agreement between the deep learning algorithm and experienced mammographers, this algorithm has the potential to standardize and automate routine breast density assessment.”

The deep learning model was implemented in routine clinical practice at Massachusetts General Hospital’s breast imaging division using more than 10,000 mammograms from January to May 2018, with the DL model achieving 94 percent agreement among MGH’s radiologists in a binary test to determine whether breasts were either heterogeneous and dense, or fatty and scattered. In addition, the model matched radiologists' assessments at 90 percent across all four BI-RADS categories.

When it came to general testing using the original dataset, the model matched the original human expert interpretations at 77 percent across four BI-RADS categories and, in binary tests, matched the interpretations at 87 percent.

“MGH is a top breast imaging center with high inter-radiologist agreement, and this high quality dataset enabled us to develop a strong model,” says Adam Yala, second author and a doctoral student in MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL).

“Our motivation was to create an accurate and consistent tool, that can be shared and used across healthcare systems,” Yala adds.

Going forward, researchers want to scale the deep learning model into hospitals besides MGH.

“Building on this translational experience, we will explore how to transition machine learning algorithms developed at MIT into clinics benefiting millions of patients,” says Regina Barzilay, senior author and the Delta Electronics Professor at CSAIL and the Department of Electrical Engineering and Computer Science at MIT.

“This is a charter of the new center at MIT—the Abdul Latif Jameel Clinic for Machine Learning in Health at MIT—that was recently launched, and we are excited about new opportunities opened up by this center,” adds Barzilay.

More for you

Loading data for hdm_tax_topic #better-outcomes...