For women with dense breasts undergoing screening mammography, the comparative effectiveness of artificial intelligence (AI) and breast ultrasound (US) alongside mammography has been a subject of ongoing investigation. This article delves into a recent study comparing the performance of mammography alone, mammography with AI assistance, and mammography supplemented by ultrasound in this specific population. The aim is to clarify the strengths and weaknesses of each approach in detecting breast cancer.
Study Overview
A retrospective analysis was conducted on data from asymptomatic women with dense breasts who underwent screening mammography and supplemental handheld whole-breast ultrasound at a primary healthcare center between January 2017 and December 2018. The study involved five breast radiologists who independently reviewed mammograms, first without AI and then with AI assistance, recording their recall decisions in both scenarios. The results of mammography combined with ultrasound were also collected. A dedicated breast radiologist further reviewed mammography findings to confirm lesion identification. The diagnostic accuracy of each method was evaluated against a reference standard of histologic examination and 1-year follow-up data.
Key Findings
The study, involving 5707 women with dense breasts, revealed significant differences in the performance of the three screening strategies.
Mammography with AI
Compared to mammography alone, the integration of AI demonstrated improved specificity and a lower abnormal interpretation rate (AIR). Specifically, mammography with AI showed a statistically significant higher specificity (95.3%) and lower AIR (5.0%) than mammography alone (94.3% and 6.0%, respectively). This suggests that AI can aid radiologists in reducing false-positive findings in mammography, leading to fewer unnecessary recalls for further investigation.
Mammography Plus Ultrasound
While AI enhanced mammography’s specificity, supplementing mammography with ultrasound offered a different set of advantages. Mammography plus ultrasound exhibited a significantly higher cancer detection rate (CDR) and sensitivity compared to mammography with AI. The CDR for mammography plus ultrasound was 5.6 per 1000 examinations, and the sensitivity was 97.0%, versus 3.5 per 1000 examinations and 60.6% sensitivity for mammography with AI. However, this increased cancer detection came at the cost of lower specificity (77.6%) and a higher AIR (22.9%) compared to mammography with AI (95.3% specificity and 5.0% AIR). Notably, supplemental ultrasound alone was responsible for detecting 12 additional cancers, the majority of which were early-stage (stage 0 and I).
Conclusion
In conclusion, while AI can enhance the specificity of mammography interpretation and reduce false positives in breast cancer screening for women with dense breasts, supplementing mammography with ultrasound proves to be more effective in detecting early, node-negative breast cancers that might be missed by mammography, even with AI assistance. Although mammography plus ultrasound leads to a higher rate of abnormal interpretations, the increased cancer detection rate, particularly for early-stage cancers, suggests its value as a supplemental screening tool in this population. This study underscores the importance of considering both AI and supplemental ultrasound in optimizing breast cancer screening strategies for women with dense breasts, highlighting the trade-offs between specificity and sensitivity in different approaches to breast imaging.
