Enhancing breast ultrasonography education: impact of artificial intelligence-based decision support on the performance of non-specialist medical professionals
1Department of Radiology, Severance Hospital, Research Institute of Radiological Science, Yonsei University, College of Medicine, Seoul, Korea 2Department of Radiology, Severance Hospital, Research Institute of Radiological Science, Yonsei University, College of Medicine, Seoul, Korea 3Department of Radiology, School of Medicine, Kyungpook National University, Kyungpook National University Chilgok Hospital, Daegu, Korea 4BeamWorks Inc., Daegu, Korea 5School of Computer Science and Engineering, Kyungpook National University, Daegu, Korea
Received: September 9, 2024; Accepted: December 12, 2024. Published online: December 12, 2024.
ABSTRACT
Purpose: To evaluate the educational impact of an artificial intelligence (AI)-based decision support system for breast US on medical professionals not dedicated to breast imaging.
Methods: This multi-case, multi-reader study was conducted as part of a breast US education course for healthcare providers. Educational material with American College of Radiology (ACR) Breast Imaging Reporting and Data System (BI-RADS) descriptors were provided along with corresponding AI results during education. The AI system displayed results as AI-heatmaps, AI scores and AI-provided BI-RADS assessment categories. Forty-two readers reviewed the test set for three separate sessions, the first session (S1) before education, the second after education without AI assistance (S2), and the third after education with AI assistance (S3). Area under the receiver operating characteristics curve (AUC), sensitivity, and specificity and performances were compared between sessions.
Results: Mean sensitivity [66.5% (95% confidence interval (95% CI): 59.2, 73.7) to 88.7% (84.1, 93.3), P<0.001] and AUC [0.664 (0.606, 0.723) to 0.684 (0.620, 0.748), P=0.300] were higher in S2 compared to S1. The AUC found by AI was comparable to that of the expert reader: 0.747 (0.640, 0.855) vs. 0.803 (0.706, 0.900), respectively (P=0.217). With AI assistance, the mean AUC for inexperienced readers was not significantly different from the expert reader; 0.745 (0.660, 0.830) vs. 0.803 (0.706, 0.900), respectively (P=0.120).
Conclusion: Mean AUC and sensitivity improved after AI was incorporated into breast US education and interpretation. AI systems with high-level performance for breast US can potentially be used as educational tools in the interpretation of breast US images.
Print
facebook
twitter
Linkedin
google+
Download Citation
PDF Links 
Full text via DOI
