Artificial Intelligence in Breast Cancer Detection – Results from Real Clinical Practice

Introduction

Artificial intelligence appears to be a promising independent reader of screening radiodiagnostic images, although there are still few prospective studies addressing this issue. Therefore, Stockholm hospital focused on the use of AI for breast cancer detection in real clinical practice.

Methodology and Study Progress 

ScreenTrustCAD is a prospective population-based study with paired evaluators to determine noninferiority, involving nearly 56 thousand women without breast implants aged 40-74 years, participants in population screening within the catchment area of Sankt Göran hospital. 

The primary outcome was screening-detected breast cancer within 3 months after mammography, and the primary aim of the analysis was to assess the noninferiority of double reading by 1 radiologist (R) and AI compared to the standard double reading by 2 radiologists. Further, the analysis included AI alone and triple reading (2R + AI) compared to double reading by 2R.

Results

Between April 1, 2021, and June 9, 2022, 58,344 women underwent regular mammographic screening, 55,581 of whom were included in the study. 

• Breast cancer was diagnosed in 269 (0.5%) women based on initial positive double reading of images.
• Double reading 1R + AI was not inferior to double reading 2R: 261 (0.5%) vs. 250 (0.4%) detected cases; relative ratio 1.04 (95% confidence interval [CI] 1.00-1.09). 
• Single AI reading compared to 2R: 246 (0.4%) vs. 250 (0.4%) detected cases; relative ratio 0.98 (95% CI 0.93-1.04) and triple reading (2R + AI): 269 (0.5%) vs. 250 (0.4%) detected cases; relative ratio 1.08 (95% CI 1.04-1.11) was also noninferior compared to standard double reading 2R. 

Discussion and Conclusion

The results of this study showed that the use of double reading in the "AI + radiologist" model for breast cancer detection is comparable to double reading by two radiologists and can even lead to a 4% increase in disease detection. This was contributed by the combination of two factors – AI has sufficiently sensitive detection capability, while the consensus second reader can exclude false positive AI assessments.

The study suggests the advantage of controlled AI implementation into the screening mammography process with emphasis on risk management and monitoring effectiveness in real clinical practice.

(lexi)

Source: Dembrower K., Crippa A., Colón E. et al. Artificial intelligence for breast cancer detection in screening mammography in Sweden: a prospective, population-based, paired-reader, non-inferiority study. Lancet Digit Health 2023; 5 (10): e703–e711, doi: 10.1016/S2589-7500(23)00153-X.