In a win for artificial intelligence, a study conducted at Baptist Health's Boca Raton (Fla.) Regional Hospital revealed the possibility of detecting breast cancer up to two years sooner using AI compared with human analysis of mammograms alone, Kathy Schilling, MD, medical director of the healthcare system's Christine E. Lynn Women's Health and Wellness Institute, told Becker's.
Using 3D mammogram scans from 54,000 patients from 2018 to 2020 and 48,000 patients from 2020 to 2022, the study compared breast cancer detection rates between human analysis and AI analysis. Dr. Schilling pointed to COVID-19 safety protocols that kept women from getting tested as the reason fewer mammograms were performed in the latter time period.
The breast cancer detection rate at the institute increased by 23 percent, according to study results — a finding Dr. Schilling called "huge."
Using AI to scan 3D exams in the study helped her team find "the tiniest of cancers possibly as early as two years sooner than they would have been diagnosed without AI," she said. "Every single day, perhaps multiple times a day, we're seeing things that refocus our attention — small things that could have easily been overlooked."
Dr. Schilling said there was no increase in recall and fewer false positives as a result of AI analysis in the study. "If we recalled patients, they were more likely to have something important compared to prior years. These are unbelievable, unexpected findings."
Digital mammography provides radiologists with four images per examination while 3D technology offers 250. "Consider that a radiologist who is reading 100 digital mammograms a day is looking at 400 images. If you're still reading 100 cases but you're using 3D mammography, you're going to see 45,000 images a day," Dr. Schilling said. "It can be totally overwhelming to the radiologist and can cause burnout and a lot of fatigue."
Findings of the study were presented at the European Conference of Radiology in Vienna in March. Dr. Schilling is planning on doing a second, more in-depth study to determine the types, sizes, histopathologies and what stages the cancers can be found with AI, as compared to human analysis, before her team submits the research for publication.
"We are all dedicated breast radiologists with fellowship training. All we do is read breast imaging. And we have an average of 22 years of experience reading breast images. We know what we are doing," she said. "AI helps us improve efficiency and decrease reading time. It's a backup to give us more confidence."
Baptist Health is based in Coral Gables, Fla.