AI risk-based algorithms could help detect the most cancers in times of reduced capacity like the pandemic when mammogram numbers are limited
Risk-based AI algorithms help detect cancer from limited number of mammograms
March 30, 2021
by John R. Fischer
, Senior Reporter
Because of COVID-19, mammograms have become limited in number from the cancellation and postponement of non-urgent services. This has made detecting breast cancer and triaging patients more difficult.
“Facilities vary in how they triage patients during the initial shutdowns and how they are triaging women as they work through backlogs. Some are scheduling based on the ordering of original appointments. Others are prioritizing diagnostic mammograms over screening mammograms, but not looking at other risk factors,” Dr. Diana Miglioretti, professor and division chief of biostatistics at UC Davis Department of Public Health Sciences, and affiliate investigator with Kaiser Permanente Washington Health Research, told HCB News.
Following extensive research in one of the largest studies of its kind, Miglioretti and her colleagues found a risk-based algorithm that can maximize cancer detection efficiently. The algorithm takes into account clinical indication, breast symptoms, breast cancer history and age. It was used to assess nearly two million mammograms of nearly 900,000 individuals taken at more than 90% of radiology facilities between 2014 and 2019.
Among mammograms that showed “very high” and “high” cancer detection rates, 12% were among 55% of all detected cancers. For those labeled with “very low” cancer detection rates, 44% accounted for 13% of all detected cancers. The findings suggest that rather than using a non-risk-based approach, breast screening practices should triage individuals most likely to have cancer during the pandemic, as it may be the best approach for detecting the most cancers despite limited availability of mammograms.
To increase the number of available mammograms during the pandemic, Miglioretti says facilities “should reach out to women who are overdue for their mammogram and let them know it is safe to come in,” as some may be hesitant to go to a medical facility while the pandemic is still going on.
She adds that vendors should “develop software to easily implement risk-based algorithms such as ours by using information already collected in the electronic medical records.”
The findings were published in JAMA Network Open.