ĐÁNH GIÁ VAI TRÒ HỖ TRỢ CỦA AI TRONG SÀNG LỌC VÀ CHẨN ĐOÁN UNG THƯ VÚ TRÊN X QUANG TUYẾN VÚ

Abstract

Purpose: evaluate the role of AI in screening and diagnosing breast cancer using mammography. This study used BIRADS scores to contrast AI-assisted radiologists with conventional approaches. Methods and materials: prospective descriptive study of 1151 women earmarked for mammography. Initial mammograms were interpreted by experienced breast radiologists (reader 1) blind to AI outputs. Simultaneously, AI systematically generated scores from 1 to 100. Mammograms were re-evaluated by a distinct radiologist (reader 2), blind to the primary assessment. Results: sensitivities of reader 2 and reader 1 were compared via ROC curve using BI-RADS scores. With reader 1 as a reference, reader 2 exhibited an AUC of 75,9% (95% CI 71,6 – 80,1%) and a sensitivity of 61,8%. When reader 2 served as the reference, reader 1 reported an AUC of 73,4% (95% CI 69,2 – 77,6%) with a sensitivity of 56,9%. When compared in the group with pathology, the reader 2 and reader 1 had AUC values ​​of 83,6% and 79,9%, respectively. This demonstrates the enhanced sensitivity with AI-assistance in breast cancer screening. Conclusion: Incorporating AI assistance into mammography has demonstrated increased sensitivity and ability to diagnose breast cancer.