Overview

An AI algorithm designed for breast cancer screening shows promise in enhancing the effectiveness of digital breast tomosynthesis (DBT), potentially reducing the rate of interval cancers by up to one-third, as reported in a study published in Radiology, the journal of the Radiological Society of North America (RSNA).

Understanding Interval Breast Cancers

Interval breast cancers are those diagnosed between regular screening mammograms and often have worse outcomes due to their aggressive nature and rapid growth. DBT, or 3D mammography, improves the visualization of breast lesions, particularly in women with dense breast tissue. However, long-term data on patient outcomes are still limited, especially in facilities that have only recently adopted DBT.

Study Insights

Dr. Manisha Bahl, the study’s lead author and breast imaging division quality director at Massachusetts General Hospital, explained:

• The interval cancer rate is often used as a surrogate marker for breast cancer-related mortality due to the lack of long-term data.
• Reducing this rate is believed to lower breast cancer-related morbidity and mortality.

Research Findings

In a retrospective analysis of 1,376 cases, the research team examined 224 interval cancers in women who had undergone DBT screening. The AI algorithm, known as Lunit INSIGHT DBT v1.1.0.0, successfully localized 32.6% (73 out of 224) of previously undetected cancers.

Dr. Bahl noted:

• Nearly one-third of interval cancers were detected and accurately localized by the AI algorithm on mammograms that radiologists had interpreted as negative.
• This highlights the potential of AI as a valuable second reader in breast cancer screening.

Significance of the Study

This research may be the first to specifically investigate AI’s role in detecting interval cancers during DBT screenings. While previous studies have focused on AI’s effectiveness in two-dimensional digital mammography, this study emphasizes its application in DBT.

Methodology

To ensure accurate sensitivity assessment, the research team conducted a lesion-specific analysis, which credits the AI algorithm only when it correctly identifies and localizes the exact cancer site. This contrasts with exam-level analyses that may inflate sensitivity by giving credit for any positive exam, regardless of accuracy.

Conclusion

The findings suggest that AI may preferentially detect more aggressive or rapidly growing tumors, indicating its potential to improve breast cancer screening outcomes. Dr. Bahl concluded:

• The AI algorithm can retrospectively detect and localize nearly one-third of interval breast cancers, supporting its integration into DBT screening workflows.
• The real-world impact of AI will depend on its adoption and validation across various clinical settings.