Using an artificial intelligence (AI)–integrated workflow, DeepHealth, in computer-aided detection of breast cancer from digital breast tomosynthesis exams found 21.6% more cases than the usual standard of care, according to findings from the AI-Supported Safeguard Review Evaluation (ASSURE) study published in Nature Health.
This study is the largest real-world analysis of AI-based breast cancer screenings yet conducted in the United States, according to its authors.
“Beyond the remarkable results, what sets this research apart is its scale, diversity, and real-world applicability,” stated Howard G. Berger, MD, President and Chief Executive Officer of RadNet, an outpatient diagnostic imaging service; DeepHealth is the AI subsidiary of RadNet. “There has never been a similar study of this size in the United States—much less one with such a diverse patient population—that examines the patient impact and efficacy of AI-assisted breast cancer screening.”
Study Methods
AI-enhanced screening methods have already shown benefits for improving detection of early breast cancer, particularly in European studies.
In ASSURE, researchers tested the AI-driven workflow on 208,891 breast cancer screening scans. Women were screened using digital breast tomosynthesis with an AI-driven protocol using DeepHealth, which has been cleared by the FDA as a computer-aided detection and diagnosis software. As a part of the company's Enhanced Breast Cancer Detection Program, with the safeguard review workflow of the AI tool, women with scans suspicious of malignancies could be sent for a second breast imaging expert review.
The performance of the AI-powered workflow was compared with the performance of 3D mammography alone on 370,692 scans.
Overall, the study population of the ASSURE study included more than 579,000 women from 109 community-based imaging sites in California, Delaware, Maryland, and New York. The population also included more than 150,000 Black women.
“Unlike many academically focused studies, these screenings took place at community imaging centers, where most women get their mammograms,” said co-study author Gregory Sorensen, MD, Chief Science Officer at RadNet.“To avoid potential selection bias, the AI-enabled workflow was provided to all patients at no additional charge during the study period.”
Key Study Findings
The AI-driven workflow demonstrated a cancer detection rate 21.6% higher than with standard screening methods (5.6 vs 4.6 per 1,000 women scanned). The recall rate was 11.1% with the AI-driven workflow vs 10.6% with standard breast cancer screenings; and the positive predictive value was 5% vs 4.4%, respectively.
No disparities in detection rate, recall rate, or positive predictive value were noted across racial and density subpopulations with the DeepHealth AI-driven workflow. Women with dense breast tissue had a 22.7% increase in their cancer detection rate with the AI tool and Enhanced Breast Cancer Detection Program as compared with 3D mammography.
"These real-world findings demonstrate how AI can improve access to specialist-level care for women, no matter where they live. When breast cancer is found early, women have far more options for care,” Dr. Sorensen added.
Disclosure: For full disclosures of the study authors, visit nature.com.
