“These findings support AI-enhanced slab reconstruction as a workflow-efficient DBT approach without compromising reader performance. Unlike traditional slabbing methods such as (maximum intensity projection), which may reduce the visibility of small amorphous calcification clusters or subtle architectural distortions, 3DQuorum uses AI-driven, spatially weighted slabbing to preserve clinically important details while still reducing the number of image slices,” noted lead studv author C.M. DeBenedectis, MD, a professor of radiology at the UMass Chan Medical School in Worcester, Massachusetts, and colleagues.
Three Key Takeaways
• Thicker AI slabs can shorten DBT reading time without sacrificing diagnostic accuracy. In this 15-reader, 391-case study, 6 mm AI-generated slabs (3DQuorum) from digital breast tomosynthesis (DBT) were non-inferior to standard 1 mm slices for overall diagnostic performance (mean AUC 80.9 percent vs. 78.9 percent). Mean reading time fell from 61.9 to 54.5 seconds per case. That is about 7 seconds, which is modest per case but could add up over a high-volume screening list.
• Unlike conventional MIP slabs, AI-enabled slabs from DBT may help preserve early disease findings. Traditional slabs with maximum intensity projection (MIP) may obscure small amorphous calcification clusters and subtle architectural distortion that may be signs of ductal carcinoma in situ (DCIS) or early invasive cancer. Through the use of spatially weighted slabs, the 3DQuorum software may retain these details while still reducing the number of slices to review, according to the study authors.
• Results need confirmation in real screening populations. The cohort was cancer-enriched (106 of 391 cases were biopsy-proven cancers), so performance and time savings may differ in routine screening where cancer prevalence is far lower. The exploratory breast density analysis was also underpowered so it remains unclear whether the benefit of the software holds equally in dense breasts. Prospective, screening-population data would be needed before relying on the utility of the aforementioned AI software.
The researchers emphasized the non-inferior findings that the 3DQ software had in comparison to conventional DBT slices for recall cases involving calcification-only cancers, and the potential impact in facilitating early detection.
“The finding of maintained performance for calcification-only cancers is particularly important because calcifications may represent DCIS or early invasive disease and may be vulnerable to reduced conspicuity with thicker slab reconstructions,” added DeBenedectis and colleagues.
(For related content, see “Mammography Study Shows Impact of AI-Powered Slab Reconstruction with DBT,” “Large Mammography Study Shows AI-Aided DBT Bolsters Cancer Detection Without Increasing Recall Rate” and “Mammography Study: Can Slab Reconstruction Technology Reinvent Efficiency with DBT?”)
In regard to study limitations, the researchers acknowledged a cancer-enriched cohort and that exploratory analysis looking at breast density was underpowered.
