ObjectiveTo analyze the results of the preliminary application of the breast cancer ultrasound intelligent diagnostic system (DeepBC) in primary-level breast cancer screening. MethodsWomen who underwent breast ultrasound screening at Shifang Maternal and Child Health Hospital in Sichuan Province and at Wanzhou Maternal and Child Health Hospital in Chongqing from January 2023 to August 2024 were retrospectively included. DeepBC performed an independent risk assessment using static breast ultrasound (BUS) lesion images acquired by sonographers and did not independently detect lesions. The distributions of Breast Imaging Reporting and Data System (BI-RADS) categories and risk-stratification results between DeepBC and sonologists were compared. BI-RADS categories 1–3 were defined as low risk, category 4 or higher as high risk, and category 0 or indeterminate results as uncertain. The agreement rate and patterns of risk upgrading and downgrading between DeepBC and sonographers were evaluated. ResultsA total of 16 750 women were included, and BUS detected 6 812 breast lesions in 4 496 participants. Sonologists classified 5 932 lesions (87.08%) as low risk, 863 (12.67%) as high risk, and 17 (0.25%) as uncertain; the corresponding numbers classified by DeepBC were 2 517 (36.95%), 4 104 (60.25%), and 191 (2.80%), respectively. Three-category agreement analysis of all 6 812 lesions yielded a Cohen Kappa of 0.135 [95%CI (0.115, 0.155)], with an overall agreement rate of 47.94% (3 266/6 812). Of the 5 932 lesions initially classified as low-risk by sonologist, DeepBC upgraded 3 308 (55.77%) to high-risk. Conversely, among the 863 high-risk lesions classified by sonologists, DeepBC concurred on 791, yielding a high-risk agreement rate of 91.66%. In a subgroup of 3 781 hypoechoic lesions, a three-class agreement analysis was performed. The Cohen Kappa between the DeepBC and sonologists was 0.215 [95%CI (0.188, 0.242)], with an overall agreement rate of 53.45% (2 021/3 781). Specifically, within this subgroup, DeepBC upgraded 1 636 (54.95%) of the 2 977 sonologist-defined low-risk lesions, and for the 798 lesions originally deemed high-risk by sonologists, DeepBC matched 740, corresponding to a concordance rate of 92.73%. ConclusionsDeepBC shows a high concordance with BUS performed by sonologists in the classification of high-risk lesions; however, the overall agreement between the two in risk stratification is low, and DeepBC frequently upgrades lesions classified as low risk by sonologists. Because complete histopathological or long-term imaging reference standards were unavailable in this study, these findings primarily reflect differences in risk stratification between the two approaches and cannot be interpreted as evidence of the diagnostic accuracy of DeepBC. Further studies incorporating histopathological confirmation, standardized imaging follow-up, and multicenter external validation are needed to evaluate its clinical value.