New research suggests that adjunctive AI with digital breast tomosynthesis (DBT) facilitates better overall detection of breast cancer, invasive breast cancer and lobular breast cancer in comparison to DBT interpretation by unaided breast radiologists.
For the retrospective study, recently published in the Journal of the American College of Radiology, researchers compared breast cancer detection rates between 54,440 unaided DBT assessments and 48,742 DBT evaluations with adjunctive AI (ProFound AI, iCAD/RadNet). The study authors noted 339 true positive cases in the unaided evaluation cohort and 369 true positive assessments in the adjunctive AI cohort.
The researchers found that adjunctive AI resulted in a 22 percent higher breast cancer detection rate (CDR), noting a 7.57 CDR per 1000 DBT exams for AI-assisted evaluations in contrast to 6.23 CDR per 1000 for unaided DBT interpretation.
The study authors noted a 7.82 percent higher detection of breast cancer detection with adjunctive AI in women with dense breasts (44.99 percent vs. 37.17 percent). Adjunctive AI also facilitated a 26 percent higher detection rate for invasive breast cancer (5.83 per 1000 vs. 4.63 per 1000) and a greater than twofold improved detection of lobular breast cancer (0.98 per 1000 vs. 0.44 per 1000).
“Interpretation of screening DBT exams by dedicated breast radiologists with concurrent use of AI resulted in an increase in CDR, invasive and lobular cancer detection rates, an increase in cancers detected in dense breasts and a decrease in mean invasive size and stage,” noted lead study author Kathy J. Schilling, M.D., FACR, the medical director of the Christine E. Lynn Women’s Health and Wellness Institute in Boca Raton, Fla., and colleagues.
There was no significant difference in recall rates between the use of adjunctive AI with DBT (6.96 percent) and unaided interpretation of DBT screening exams (6.97 percent), according to the study authors.
Three Key Takeaways
- Adjunctive AI improves overall and invasive cancer detection without increasing recalls. AI-assisted DBT increased overall cancer detection by 22 percent (7.57 vs. 6.23 per 1000 exams) and invasive cancer detection by 26%, with no significant change in recall rates, suggesting improved diagnostic yield without added false positives.
- Enhanced detection in dense breasts and lobular cancers. AI use was associated with higher detection in women with dense breasts (44.99 percent vs. 37.17 percent) and more than doubled detection of lobular carcinoma (0.98 vs. 0.44 per 1000), addressing traditionally challenging diagnostic subgroups.
- Earlier-stage and smaller invasive cancers detected, without increased DCIS diagnosis. AI-assisted DBT identified smaller invasive tumors (mean 10.74 mm vs. 12.16 mm) and more T1-stage cancers (70.7 percent vs. 63.1 percent), without increasing DCIS detection, suggesting improved early detection without evidence of over diagnosis.
The researchers also noted that use of adjunctive AI also led to a reduced mean size of detected invasive cancers (10.74 mm vs. 12.16 mm as well as increased detection of T1 stage cancers (70.7 percent vs. 63.1 percent) without a significant change in ductal carcinoma in situ (DCIS) detection (23 percent vs. 25.7 percent).
“With no change in the rate of DCIS diagnosis, it suggests that the use of AI can improve radiologists’ performance without over diagnosing. Detection of smaller cancers with the use of AI may change treatment options and overall outcomes for women with screen-detected cancers,” pointed out Schilling and colleagues. “Localized tumor burden has the potential for breast conserving surgery options without lymph node surgery and with improved radiation or endocrine therapy options.
(Editor’s note: For related content, see “Mammography Study: Can Slab Reconstruction Technology Reinvent Efficiency with DBT?,” “Mammography Study Shows Advantages of DBT Guidance for Breast Biopsies” and “FDA Clears AI-Powered Triage Platform for Digital Breast Tomosynthesis.”)
Beyond the inherent limitations of a retrospective study, the authors acknowledged the use of one mammography vendor and one AI software. The researchers also conceded that the study findings, drawn from a largely White, non-Hispanic cohort, may not be applicable to broader populations.
