Can AI have an impact in reducing interval breast cancers?
In the randomized controlled Mammography Screening with Artificial Intelligence (MASAI) trial, recently published in the Lancet, researchers compared the use of adjunctive AI (Transpara, version 1.7.0., ScreenPoint Medical) for mammography triage to double reading by radiologists without AI in 105,915 women (approximate median age of 53).
The study authors found that that adjunctive AI facilitated a 12 percent reduction in the interval cancer rate and a 16 percent decrease in the rate of invasive interval breast cancers.
Among the women with detected interval cancers, the researchers noted that the adjunctive AI cohort had less Luminal B breast cancer (23 vs. 30 cases), triple-negative breast cancer (12 vs. 16) and HER2-positive, ER-positive breast cancer (5 vs. 7). Overall, there was a 27 percent reduction in the number of invasive cases involving non-Luminal A molecular subtypes with the use of adjunctive AI in screening (43 vs. 59 cases).
“The interval cancer rate is an important indicator of screening efficacy, and although this study was not powered to show superiority, these results suggest a potential clinical benefit of earlier detection of clinically relevant breast cancer, which might enable less aggressive treatment and improved prognosis,”noted lead study author Jessie Gommers, MSc, who is affiliated with the Department of Medical Imaging at Radboud University Medical Center in Nijmegen, Netherlands, and colleagues.
Three Key Takeaways
• Adjunctive AI reduced interval cancers without sacrificing specificity.
In the MASAI randomized trial, AI-supported mammography screening achieved a 12 percent reduction in interval cancer rates and a 16 percent reduction in invasive interval cancers, while maintaining equivalent specificity (98.5 percent) compared with standard double reading.
• AI preferentially reduced biologically aggressive interval cancers.
Screening with adjunctive AI was associated with fewer non-Luminal A invasive cancers, including Luminal B, triple-negative, and HER2+/ER+ subtypes, amounting to a 27 percent reduction in invasive non-Luminal A interval cancers, suggesting earlier detection of clinically relevant disease.
• Sensitivity gains were consistent across density and age, including dense breasts. AI improved overall sensitivity (80.5 percent vs. 73.8 percent) and showed a notable benefit in women with extremely dense breasts (+11.1 percent sensitivity), supporting its role as a screening adjunct in populations where mammography performance is traditionally limited.
The researchers also pointed out higher sensitivity rates with adjunctive AI regardless of age or breast density category. For women with extremely dense breasts, adjunctive AI facilitated an 11.1 percent increase in sensitivity (71.1 percent vs. 60 percent). In addition to the higher overall sensitivity rate with adjunctive AI (80.5 percent vs. 73.8 percent, the study authors noted equivalent specificity (98.5 percent).
“ … The MASAI trial showed consistently more (favorable) outcomes with AI-supported mammography screening compared with standard double reading without AI, including the primary outcome of interval cancer rate, showing non-inferiority, and fewer interval cancers with (unfavorable) characteristics,” added Gommers and colleagues.
(Editor’s note: For related content, see “Large Mammography Study Affirms Value of AI in Breast Cancer Detection,” “Mammography Study Shows Elevated Future Breast Cancer Risk with Initial Concordance of Radiologist and AI Interpretation” and “Predicting Interval Breast Cancer Risk: Can a Mammography Deep Learning Model Have an Impact?”)
In regard to study limitations, the authors acknowledged the use of one mammography vendor, one AI software and one round of screening. They also noted a lack of race and ethnicity data for the cohort, which was drawn from four facilities in Sweden.
