[Image from maf04 on Flickr]
A team of researchers from MIT’s CSAIL, Massachusetts General Hospital and Harvard Medical School developed an AI system with machine learning that can predict if high-risk lesions on needle biopsies found during a mammogram will turn out to be cancerous after surgery.
High-risk lesions that appear on mammograms and have abnormal cells when tested using a needle biopsy often are a cause of false positives in breast cancer. A patient goes through surgery to have the lesion removed, but 90% of the time, the lesion was benign. The researchers are using AI to eliminate that.
The team of researchers tested their AI system on 335 high-risk lesions. It correctly diagnosed 97% of the breast cancers as malignant and reduced surgeries of benign lesions by more than 30%.
“Because diagnostic tools are so inexact, there is an understandable tendency for doctors to over-screen for breast cancer,” Regina Barzilay, MIT’s Delta Electronics professor of electrical engineering and computer science and breast cancer survivor, said in a press release. “When there’s this much uncertainty in data, machine learning is exactly the tool that we need to improve detection and prevent over-treatment.”
Machine learning models developed by the researchers are trained from more than 600 existing high-risk lesions. The models are trained to identify patterns in different data points like demographics, familial history, past biopsies and pathology reports.
“To our knowledge, this is the first study to apply machine learning to the task of distinguishing high-risk lesions that need surgery from those that don’t,” Constance Lehman, a collaborator on the study and chief of the breast imaging division at Massachusetts General Hospital department of radiology, said. “we believe this could support women to make more informed decisions about their treatment, and that we could provide more targeted approaches to healthcare in general.”
This year alone, there is expected to be 252,710 new cases of breast cancer in the U.S. and 40,610 breast cancer-related deaths, according to the American Cancer Society. One in eight women will have breast cancer in their lifetime.
Mammograms are general practice for detecting and diagnosing the disease. When mammograms come back with suspicious lesions, a patient has to undergo a needle biopsy to determine if the lesion is cancerous. According to the researchers, 70% of the lesions end up being benign, 20% are malignant and 10% are high-risk.
Traditionally, some doctors perform surgery on any high-risk lesions. Other doctors perform surgery on lesions that have higher cancer rates like atypical ductal hyperplasia (ADH) or lobular carcinoma in situ (LCIS).
Oftentimes, the surgical procedures are time-consuming and expensive and leave patients with scarring. And then, surgeons removing lesions could miss cancers other than ADH and LCIS.
“The vast majority of patients worth high-risk lesions do not have cancer, and we’re trying to find the few that do,” said Manisha Bahl, a doctor in Massachusetts General Hospital’s department of radiology. “In a scenario like this, there’s always a risk that when you try to increase the number of cancers you can identify, you’ll also increase the number of false positives you find.”
The AI system developed by the team uses a method called random-forest classier and results in fewer unnecessary surgeries compared to traditional methods.
“This work highlights an example of using cutting-edge machine learning technology to avoid unnecessary surgery,” Marc Kohli, director of clinical informatics in the department of radiology and biomedical imaging at the University of California at San Francisco, said. “This is the first step toward the medical community embracing machine learning as a way to identify patterns and trends that are otherwise invisible to humans.”
Massachusetts General Hospital radiologists are expected to begin using the AI model in their clinical practices over the next year.
“In the past, we might have recommended that all high-risk lesions be surgically excised,” said Lehman. “But now, if the model determines that the lesion has a very low chance of being cancerous in a specific patient, we can have a more informed discussion with our patient about her options. It may be reasonable for some patients to have their lesions followed with imaging rather than surgically excised.”
The research team plans to continue perfecting the model in the time being.
“In future work, we hope to incorporate the actual images from the mammograms and images of the pathology slides, as well as more extensive patient information from medical records,” Bahl said.
The research was published in the journal Radiology.
