Study: Radiologist using artificial intelligence detected 20% more cancers than those read only by radiologists without AI help
By Deborah Jeanne Sergeant
A study published by Swedish researchers in August indicates that artificial intelligence (AI) can safely augment and even improve radiologists’ work in detecting breast cancer.
The study is the first trial that compares AI-assisted mammogram readings with human-only mammogram readings.
The radiologist using AI detected 20% more cancers than those read only by two radiologists without AI help.
Using AI didn’t appear to increase the number of false positive mammograms, which is when an abnormality is noted but not actually present.
In addition to greater efficacy, using AI also reduces the work hours needed to read mammograms by 44% because AI replaces the work of one radiologist (some hospitals require that two radiologists are required to read each mammogram). This can be particularly helpful in health systems with few radiologists, such as in rural areas.
According to the American College of Radiology, 82% the 20,970 radiologists providing patient care are age 45 and older and 53% are 55 and older. The demand for radiologists is rising as the aging Baby Boomers require more care. The Bureau of Labor Statistics states that the need for new radiologists is expected to rise by 6%, “faster than average” between 2022 and 2032 compared with all other occupations. Enlisting the help of AI can help mitigate this rise in demand.
“Early research appears to show that AI-assisted mammograms may be an important tool to help radiologists detect breast cancer,” said Susan Brown, registered nurse and senior director of health information and publications at Susan G. Komen headquarters in Dallas. “It is hoped that AI-assisted mammograms will increase accuracy and efficiency.”
Brown noted that false positives and false negatives could be an issue with AI-read mammograms but hopes that additional research will confirm the early studies on AI assisted mammograms.
Nancy Wayne, marketing administrator at Rochester-based Elizabeth Wende Breast Care, wants to see additional research to show that AI plus a radiologist is just as good as two radiologists.
“AI is a wonderful tool,” she noted. “We’re still early on in the development of AI. There are multiple products out there currently. We’re still not quite able to say one product fits all. There are multiple companies with proprietary products. We’re hopeful where at some point the AI has to fulfill certain criteria to be appropriate. I’m not sure at this point we’re there. It is promising.”
“AI is everything in imaging,” said Avice O’Connell, director of the UR Medicine breast imaging program. “But it’s not going to take our jobs. AI is something we’ve been using for more than 20 years in mammography. In 1998 the first computer aided detection the computer-scanned images and highlighted things that could be cancer. There’s still that kind of computer-aided detection. There are usually four marks on every exam, so a human still has to look at it. It’s just an aid to make sure we’re not missing things.”
O’Connell views AI as a means to screen the “easy” cases more quickly so radiologists can pay more attention to more challenging cases. But in either, a human must be involved; AI is only a tool but doesn’t make any decisions about health.
“The patient makes the ultimate decision whether or not to have the biopsy,” O’Connell said. “We’d still follow up the patient in six months.”
O’Connell sees machine learning as the next threshold of mammography advancement, meaning that AI technology reads more and more mammograms; it becomes more skilled at catching abnormal findings. She also thinks that AI-assisted mammography will help in areas of the world where few specialists exist.
“The vast majority of breast imaging around the world is not read by breast experts,” she said. “AI has huge value in every country.”