- Researchers are reporting that artificial intelligence programs were more accurate in predicting breast cancer risk than traditional methods.
- They said the programs could help with earlier diagnosis and better preventive measures.
- Experts say artificial intelligence will become a bigger part of healthcare in the future.
Artificial intelligence (AI) programs performed better at predicting five-year breast cancer risk than traditional models, according to a study published today in Radiology, a journal of the Radiological Society of North America (RSNA).
The researchers used data from negative 2d mammograms performed at Kaiser Permanente Northern California in 2016.
The scientists screened 324,009 women and chose 13,628 for analysis. In addition, 4,584 from the eligibility pool diagnosed with breast cancer within five years of the original 2016 mammogram remained in the study.
The scientists followed the participants until 2021.
An AI program evaluated the mammograms and divided the results into three categories:
- Interval cancer risk – incident cancers diagnosed between zero and one years
- Future cancer risk – incident cancers diagnosed between one and five years
- All cancer risk–incident cancers diagnosed between zero and five years
The researchers used five AI algorithms, including two used by researchers and three that are commercially available.
The scientists compared risk scores to each other and the Breast Cancer Surveillance Consortium (BCSC).
Dr. Richard Reitherman, a radiologist and medical director of breast imaging at MemorialCare Breast Center at Orange Coast Medical Center in California, explained the factors used to calculate the risk of breast cancer.
He noted the risk is frequently calculated using the BCSC, which operates primarily five elements:
- A woman’s age
- Family history of breast cancer in a first-degree relative (mother, sister, daughter)
- Mammographic breast density
- History of benign breast biopsies.
“Many computer risk calculators are available to estimate a woman’s risk based on these factors,” Reitherman, who was not involved in the study, told Medical News Today. “Relevant to the current publication, the risk of a woman being diagnosed with breast cancer in the next five years is a standard metric.”
The scientists noted that all five AI algorithms performed better than the BCSC in predicting breast cancer risk at zero to five years.
Some algorithms predicted patients at high risk of interval cancer, which is often aggressive and might require a second mammogram or additional imaging and screening.
Other algorithms could predict future cancer risk up to five years when the mammography detected no cancer.
When predicting cancer risk in the highest 10% risk group, researchers reported that AI predicted up to 28% of cancers while the BCSC method predicted 21%.
“This study is particularly interesting because all but one of the AI models examined were designed to detect the presence or absence of breast cancer in a specific mammogram, not to predict a woman’s future risk of developing cancer,” explained Dr. Laura Heacock, a breast radiologist at NYU Langone Perlmutter Cancer Center in New York who was not involved in this study but has authored other
“This is noteworthy because conventional breast cancer risk models employed by physicians and providers can require extensive information such as family history, ethnicity, prior breast biopsies, pregnancy, and hormone usage,” she told Medical News Today.
“Despite relying solely on a one-time mammogram examination, these AI models outperform the BCSC model in identifying women more likely to develop cancer in the future,” Heacock added. “The use of AI in predicting both current and future breast cancer risk represents a powerful approach that leverages AI for individual benefit.”
“AI studies like this show that not all dense breasts are equal; specific and complex breast tissue patterns exist that predict a higher risk of breast cancer,” Heacock noted. “AI may identify patterns imperceptible to the human eye or are only visible by training on hundreds of thousands of mammograms.”
They believe AI identifies missed cancers and breast cancer features that could predict future cancer development.
“The interesting message of this article is that AI can be used to identify areas in a mammogram or other mammographic features that are not yet cancer (and therefore cannot be currently diagnosed) but may develop into cancer in the next five years,” Reitherman said. “This capability can direct appropriate more sensitive resources such as breast ultrasound or Breast MRI to be integrated into the woman’s screening management. Risk reduction management techniques such as endocrine blockade may also become more important.”
“I would absolutely be willing to use this technology if there is human regulation,” Reitherman said. “Human interface is critical – there need to be some controls.”
AI is already being used in women’s healthcare.
“Mammography has been using AI in the setting of ‘computer-aided detection, aka CAD, since [Food and Drug Administration] approval in 1998,” said Dr. Kenneth Meng, the medical director at the Center for Breast Imaging and Diagnosis at the Center for Cancer Prevention and Treatment at Providence St. Joseph Hospital in California.
“Currently, there have been some advancements in various forms of computer-aided detection, but no one is relying on these systems alone,” Meng, who was not involved the study, told Medical News Today. “I would say most mammograms are read with some form of computer-aided detection (with some markings of areas of concern detected by the software algorithm, but typically at the end of the readout). These can inform the radiologist to take a second look, but ultimately, there is much disagreement between the CAD markings and the radiologist’s final interpretation at this time.”
Experts say it seems inevitable that AI’s role in healthcare will continue to increase. It can help in many different areas, including administration, patient engagement, surgical robots, and diagnostics, according to a
As the current study notes, radiology is one area of medicine where AI seems to be well-suited – they can use thousands of images in their memory to compare to an image to determine if cancer is present or if conditions might lead to cancer.
However, several obstacles must be overcome before AI is integrated into diagnostic procedures, such as predicting future cancer. According to the report, “For widespread adoption to take place, AI systems must be approved by regulators, integrated with EHR systems, standardized to a sufficient degree that similar products work similarly, taught to clinicians, paid for by public or private payer organizations and updated over time in the field.”
Additionally, AI systems will first augment clinicians rather than replace them.
“Human interface is critical. There need to be some controls,” said Reitherman.
AI is becoming more prevalent, although it still has a long way to go.
“I am a huge proponent of AI integration into women’s medical care,” said Meng. “At expert facilities, it will still be sometime before AI shows a large advantage, but it may in the near term provide some baseline consistency or standard floor for those places that may not have access to expert radiologists.”