Application Value of Artificial Intelligence Ultrasound-Assisted Diagnosis System in the Diagnosis of BI-RADS 4 Breast Nodules
Year: 2026
Type: article
Study design: Observational Study
Abstract: To investigate the clinical application value of an artificial intelligence (AI) ultrasound-assisted diagnostic system in the diagnosis of Breast Imaging Reporting and Data System (BI-RADS) category 4 breast nodules. A retrospective analysis of 231 breast nodules from 226 patients (aged 26 to 80 years, mean 55.52 ± 12.32 years) was carried out from December 2024 to December 2025. Nodules were divided into a malignant group and a benign group according to the pathological results of biopsy or surgery. Four diagnostic approaches were compared: low-experience ultrasound physicians (15 years, US+), AI-assisted diagnosis alone (AI), and AI-assisted diagnosis combined with low-experience physicians (AI+US), all based on the BI-RADS classification system. A result of BI-RADS 4a or lower was regarded as negative, whereas 4b or above was positive. Sensitivity (SE), specificity (SP), accuracy (ACC), positive predictive value (PPV), negative predictive value (NPV) and area under the receiver operating characteristic curve (AUC) were computed with the pathological results as the gold standard. 128 of 231 nodules were malignant and 103 benign. The AI group achieved SE 82.03%, SP 80.58%, ACC 81.38%, PPV 84.00%, NPV 78.30%, and AUC 0.813 (95% CI: 0.756–0.870), exceeding the US group (all p < 0.05 vs. US). The AI ultrasound-assisted diagnostic system shows good efficacy in the diagnosis of BI-RADS 4 breast nodules, especially for small nodules.
Institution: Mongolian National University of Medical Sciences
Language: English
Topic: AI in cancer detection
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7235148855
