New Research Supports Clinical Value of ProFound AI in DBT

Two new studies presented at recent breast imaging meetings support the clinical value of ProFound AI aiding breast cancer detection in digital breast tomosynthesis.

At the Society of Breast Imaging (SBI) Symposium, Emily Conant, MD, Professor and Division Chief of Breast Imaging at the University of Pennsylvania Medical Center, presented findings from a retrospective analysis involving ProFound AI for DBT in a presentation titled “Feasibility of automated identification of low-likelihood of cancer cases in digital breast tomosynthesis screening. According to study findings, ProFound AI for DBT accurately identified 33.4 percent of normal screening DBT exams with no cancers being missed, based solely on the ProFound AI Case Score. When researchers also factored in breast density and age, ProFound AI identified 58.6 percent of normal cases with no false negatives.

“Our retrospective study demonstrates the feasibility that clinical algorithms have the potential to triage and reduce screening DBT workload by flagging normal mammograms using an AI system, and also prioritizing complex cases that are more likely to require additional review or evaluation,” said Dr. Conant. “We are pleased to have our research add to the important growing body of evidence supporting the significance and value of AI in breast screening.”  

The study was conducted to evaluate the thresholds at which the ProFound AI system could be used for triaging DBT exams to reach a minimum rate of false negatives per 1,000 screened in an enriched dataset of 506 biopsy-proven cancer cases and 1,293 non-cancer cases with 320 days of negative follow-up. A consecutive series of cases were collected from 18 sites in the United States and three sites in France.

In new data presented at the National Consortium of Breast Centers (NCBC) Annual Interdisciplinary Breast Center Conference (NCoBC), Mark Traill, MD, University of Michigan Health, presented findings from a study titled “Correlation between BI-RADS Assessment Categories and Artificial Intelligence Case Scores,” which was a winner in the “Breast Disease Diagnosis and Management” category. Dr. Traill highlighted the comparison of ProFound AI Case Scores to BI-RADS assessment categories determined by a single radiologist without using AI in a retrospective analysis. Researchers used ProFound AI on 890 consecutive DBT studies and 50 consecutive cases with biopsy-proven breast cancer detected with DBT. Results showed a strong positive correlation between a ProFound AI Case Score of less than 60 percent and patients assessed as likely to be normal (BI-RADS 1 or 2), while most of the biopsy-proven cancers had a Case Score of greater than 60 percent.

“We wanted to describe the Case Score distribution in a screening population to better understand the significance of score value as a clinical decision tool,” said Dr. Traill. “We found a very strong correlation between a Case Score of less than 60 percent and a BI-RADS score assessment of 1 or 2. Also, only 15 percent of the Case Scores were greater than 60 percent, but this group contained most of the detected cancers. As a clinical decision tool, a Case Score above 60 percent is an independent indicator of higher chance of underlying malignancy. This is very helpful in guiding the intensity of the cancer search, while improving workflow functionality and reducing stress for the reading radiologist.”

“These two studies both suggest that ProFound AI Case Scores provide valuable insights that can help clinicians more efficiently identify normal mammograms, which may directly translate to time-savings benefits,” said Michael Klein, Chairman and CEO of iCAD. “In addition, ProFound AI is clinically proven to improve radiologists’ sensitivity while simultaneously improving their specificity, which is a huge performance achievement in breast care. With the addition of these two important abstracts, research now shows how Case Scores can be used in the clinical setting to help radiologists feel more confident in their decisions about when a mammogram is normal.”