Bolstering the Performance of Breast Radiologists with AI-CAD in Mammography: A Multireader Study

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• Breast cancer is a leading global malignancy in women, with high incidence-to-mortality ratios in many Asian countries due to limited screening resources.
• Artificial intelligence (AI) offers potential solutions to challenges in mammography interpretation, such as resource constraints and manpower limitations.

Purpose of the Study:

• To evaluate the impact of AI-based computer-aided diagnosis (AI-CAD) on the diagnostic performance and interpretation time of mammograms.
• To assess the effectiveness of AI assistance for radiologists in a multinational Asian setting.

Main Methods:

• A multinational retrospective study involving 302 digital mammograms (89 malignant).
• Nine experienced breast radiologists performed interpretations in two sessions: unaided and with AI-CAD assistance.
• Diagnostic performance metrics (AUC, sensitivity, specificity) and interpretation time were compared between sessions.

Main Results:

• AI-CAD significantly improved the area under the curve (AUC) from 0.799 to 0.851 (p=0.0151).
• Specificity increased from 77.0% to 88.4% (p=0.03), with no significant change in sensitivity.
• Average interpretation time decreased significantly from 121.5 to 83.2 seconds per case (p<0.001).

Conclusions:

• AI-CAD enhances specificity and reduces reading time in mammography interpretation without compromising sensitivity.
• Findings support AI integration into breast cancer screening workflows for improved efficiency and outcomes.
• Emphasizes the continued importance of human oversight in AI-assisted clinical practice.