Clinics across the country are starting to offer patients a new service: mammograms are read not only by radiologists but also by artificial intelligence models. Hospitals and companies offering these tools tout their ability to speed up radiologists' work and detect cancer earlier than standard mammography alone.
Mammograms currently identify approximately 87 percent of breast cancers. Cancers are more likely to be missed in younger women and women with dense breasts. They can also cause false positives that require further testing to rule out cancer, or they can reveal pre-cancerous conditions that may not cause serious problems, but can still put you at risk of not treating them. cannot be predicted and may lead to treatment.
“This is not a perfect science by any stretch of the imagination,” says Dr. John Lewin, director of breast imaging at Smilow Cancer Hospital and Yale Cancer Center.
With 300,000 women diagnosed with breast cancer each year in the United States, experts are excited by the prospect of improving the accuracy of breast cancer screening. However, there are also concerns about whether these AI tools will work well for a variety of patients and whether they can meaningfully improve breast cancer survival rates.
How does AI analysis work?
Mammography contains a wealth of information about breast tissue and milk ducts. Certain patterns, such as bright white spots with jagged edges, can be a sign of cancer. In contrast, a thin white line may indicate benign calcification or may require further testing. Other patterns may be difficult for humans to distinguish from normal breast tissue.
In some cases, AI models “can see things we can't see,” said Dr. Katerina Doderzon, a radiologist who specializes in breast imaging at NewYork-Presbyterian/Weill Cornell Medical Center.
When the images are run through the AI program, the software highlights suspicious areas that require further attention by the radiologist. Some models can even score images to help busy radiologists prioritize which scans to examine first.
“I easily read 100 screening mammograms a day,” Dr. Malone said. “You can start reading what AI is saying is more complex.”
In one of the largest studies of AI mammography, a model used in Sweden improved breast cancer detection by 20%. In a trial of 80,000 women, the software detected 6 cancers per 1,000 women, while radiologists detected 5 per 1,000 women.
A 2022 Danish study showed that an AI model also reduced false positives, reducing the number of women who had to see a doctor for further testing after a suspicious area was identified on a mammogram. It means what you did.
However, it is unclear whether AI analysis will actually reduce breast cancer deaths or simply increase the number of survivors by detecting more cancers earlier. And radiologists are unsure how European study results will translate to the United States or how well the model will work with a more diverse population.
“Developing these AI tools and algorithms across different races and ethnicities requires more diverse training and testing,” Dr. Doderzon said. “AI is just a tool that learns based on what it sees.”
Some experts caution against using these tools before they have been thoroughly scrutinized, drawing comparisons to computer-aided detection techniques that were hailed in the 1980s as a way to help detect breast cancer early. Are concerned. A subsequent large-scale study found that this technology did not make mammography results more accurate.
When it comes to AI analysis of mammography, “we may not know for two to three years whether performance has degraded,” Dr. Lewin said.
There are still some things that AI still cannot do well, such as distinguishing between surgical scars and tumors. “All it takes is a human being,” says Dr. Carolyn Malone, a radiologist in the breast division of the John Theurer Cancer Center at Hackensack University Medical Center. She said radiologists, especially those trained in breast imaging, can use the patient's medical history and her own expertise to pinpoint these abnormalities.
.
Is it worth paying for AI mammography?
The Food and Drug Administration has approved approximately 24 mammography AI products. Some of these have been rolled out to patients in a small number of clinics and are being tested by other hospitals who want to see the value these tools provide before offering them to patients.
Currently, there is no billing code that radiologists can use to bill insurance companies for this technology. That means some centers may charge $40 to $100 out-of-pocket for the AI analysis and pass the cost on to patients. Other hospitals may cover the cost and provide additional analysis free of charge. Still others are holding onto the technology for research until they are more certain of the value it can provide to patients.
It will be some time before AI becomes part of daily medical care, but when it does, insurance companies will likely consider reimbursement. Until then, Dr. Doderzon said, most patients won't need his AI for mammograms, but for those who are particularly concerned about their results, the AI may provide additional peace of mind.