Objective:

To compare the diagnostic efficiencies of artificial intelligence (AI) S-Detect technique for diagnosing breast tumors using by a well-experienced ultrasound physician and a less-

experienced one and to explore the value of S-Detect technique for clinical application.

Methods:

A total of 100 patients (with 144 breast masses) who underwent breast tumor surgery in Shanghai Changning Maternity and Infant Health Hospital were included in this study and were examined by a well-experienced ultrasound physician and a less-experienced one using conventional ultrasound with Breast Imaging Reporting and Data System (BI- RADS) classification (BI-RADS ＞4A was considered as malignant) and S-Detect technique (benign nodules were diagnosed if the results of both the longitudinal and transverse were possibly benign; otherwise

malignant nodules were diagnosed) respectively. With pathological results as golden standards

the diagnostic efficiencies and consistency of conventional ultrasound and S-Detect technique were analyzed and compared.

Results:

Among 144 breast tumors

124 were benign and 20 were malignant. Area under curves (AUCs) of receiver operating characteristic (ROC) curves of conventional ultrasound of the well-experienced ultrasound physician and the less-experienced one were 0.868 and 0.690 respectively; AUCs of S-Detect technique of those physicians were 0.877and 0.893 respectively. AUC of conventional ultrasound classification of the less-experienced ultrasound physician was significantly lower than the other three (

P

＜0.01). The diagnostic results of two ultrasound physicians using conventional ultrasound were moderately consistent with intraclass correlation coefficient (ICC) as 0.736; and the diagnostic results of two ultrasound physicians using S-Detect were highly consistent with ICC as 0.928.

Conclusion:

The work experience of ultrasound physicians with different clinical working years has influence on the differential diagnosis of breast nodules

but AI technology can reduce the influence of experience on the diagnosis results. S-Detect technique is worth to be popularized clinically.