The University of Washington and University of California, Los Angeles (UCLA) researchers have developed an artificial intelligence (AI) system that may help pathologists improve the accuracy of biopsy reviewed as well as lead to better detection and diagnosis of breast cancer.
The team recently published the results of the performance of this new algorithm—apparently working nearly as accurate or better than an experienced pathologist, depending on the task.
Published in JAMA Network Open on August 9, the results originate from a 2015 study from the UW School of Medicine which found that pathologists often disagree on the interpretation of breast biopsies, which are performed on millions of women each year.
Original Study Results
The original study showed that diagnostic errors occurred for about one out of every six women who had a noninvasive type of breast cancer called “ductal carcinoma in situ.” Also, incorrect diagnoses were given in about half of the biopsy cases with abnormal cells that are associated with a higher risk for breast cancer—a condition called atypia.
Back then the scientists reasoned that AI could do a better job and hence the foundation for the present study involving UW and UCLA.
The study used a large dataset that makes it possible for the machine learning system to recognize patterns associated with cancer that are difficult for doctors to see. Post review of breast biopsy interpretation approaches, the team developed image analysis methods tailored to address specific challenges.
The team fed 240 breast biopsy images into a computer, training it to recognize patterns associated with several types of breast lesions, ranging from noncancerous and atypia to ductal carcinoma in situ and invasive breast cancer. Importantly, the team established a disciplined process to establish the correct diagnosis—a consensus among three expert pathologists.
The team compared its readings to independent diagnoses made by 87 practicing U.S. pathologists who interpreted the same cases to test the actual system. The algorithm performed nearly as well as the human pathologists in a few scenarios and outperformed humans when differentiating ductal carcinoma in situ from atypia, correctly diagnosing pre-invasive breast cancer biopsies about 89% of the time, compared to 70% to the human pathologists.
Graduate student Ezi Mercan contributed to the AI by investing a novel descriptor called “the structure feature” that was able to represent key patterns compactly for use in the AI machine learning
National Institutes of Health
Linda Shapiro, professor UW Paul G. Allen School of Computer Science & Engineering and the UW electrical and computer engineering department
Dr. Joann Elmore, Professor of Medicine, David Geffen School of Medicine, UCLA
Sachin Mehta, doctoral student UW electrical and computer engineering
Dr. Jamen Bartlett, Southern Ohio Pathology Consultants
Dr. Donald Weaver, University of Vermont