By Lynn Antonopoulos
Integrating AI with digital breast tomosynthesis (DBT) in cancer screening may enhance detection of challenging lesions while minimizing overdiagnosis, according to a Sunday morning presentation.
The talk highlighted a performance review of an AI tool used to assist radiologists in breast cancer screening and diagnosis starting in 2020.
“In particular, this tool may help breast imagers avoid overestimation of ductal carcinoma in situ (DCIS), in turn supporting more precise cancer diagnoses,” said Kathy Schilling, MD, medical director of Christine Lynn Woman’s Health and Wellness Institute at Boca Raton Regional Hospital, FL.
Dr. Schilling and her team implemented DBT AI in 2020, the same week her facility closed for six weeks due to the COVID-19 pandemic.
“We were no longer screening but only seeing patients for diagnostic studies once weekly,” Dr. Schilling said. “However, we used that to our benefit knowing all patients had symptoms and many had findings.”
The setback may have provided a jump start in realizing the capabilities of AI in cancer detection, Dr. Schilling added.
After two years of working with the AI tool, Dr. Schilling and her team sought to verify and validate their outcomes.
“I believe all programs should be evaluated after a period of time to ensure they are positively contributing to the outcomes of our practice,” she said. “I was also curious as I noticed an increase in very small cancers being identified, particularly lobular cancers. I thought this might be evidence of the impact of AI on diagnostic outcomes.”
The researchers conducted a retrospective analysis of the mammography audit data and screening cancers detected at four sites. They targeted two time periods that involved the work of nine dedicated breast radiologists: one pre-AI and the other post-AI implementation.
The researchers collected information on patient age, breast density, tumor size, staging and histopathology for all screen detected cancers.
During the first period (pre-AI), there were 54,440 exams including 339 true positives. The second period (post-AI) had 48,742 exams with 369 true positives.
AI use increased cancer detection rates from 6.23 to 7.57 per thousand exams. “We noted significant increases in detecting cancers in dense breasts, smaller tumor sizes and T1-stage invasive cancers,” Dr. Schilling said.
The detection rate for invasive cancer rose by 1.2 per thousand with an increase in lobular cancer detection from 0.44 to 0.98 per thousand screenings.
“As we identify smaller cancers, we may have the opportunity to revise our standards of treatment for our patients,” Dr. Schilling said. “We are seeing cancers perhaps years earlier than we would, were we not using AI to assist in diagnosis.”
According to Dr. Schilling, the most surprising outcome of her team’s research related to the radiologists’ levels of expertise.
“All nine fellowship-trained dedicated breast radiologists improved cancer detection without an increase in recall rate with the use of AI,” she said. “We thought we were expert readers, but found there was room for improvement. Imagine the impact AI could have on general radiologists reading DBT.”
Access the presentation, “Real World Effect of Artificial Intelligence on Histopathology and Stage in Breast Cancer Screening with Digital Breast Tomosynthesis,” (S2-SSBR01-6) on demand at RSNA.org/MeetingCentral.
© 2024 RSNA.
The RSNA 2024 Daily Bulletin is the official publication of the 110th Scientific Assembly and Annual Meeting of the Radiological Society of North America. Published online Sunday, December 1 — Friday, December 6.
The RSNA 2024 Daily Bulletin is owned and published by the Radiological Society of North America, Inc., 820 Jorie Blvd., Suite 200, Oak Brook, IL 60523.
