Real-Time Artificial Intelligence-Assisted Ultrasound System for Differentiating Ductal Carcinoma In Situ and Invasive Ductal Carcinoma

Purpose: Real-time artificial intelligence (AI)-based computer-aided detection/diagnosis (CAD) systems for breast ultrasound provide immediate probability of malignancy (POM) estimates, which may aid in differentiating ductal carcinoma in situ (DCIS) from invasive ductal carcinoma (IDC).Identification of the ultrasonographic features associated with significant POM differences has the potential to improve the clinical classification.Methods: A retrospective analysis was performed on 34 cases (DCIS, n = 12; IDC, n = 22) evaluated using a real-time AI solution (CadAI-B for breast cancer).The POM outputs and Breast Imaging Reporting and Data System (BI-RADS) categories were compared across the ultrasonographic features, including shape, margin, echo pattern, and orientation.Comparative analyses identified features showing significant POM differences between DCIS and IDC and assessed the diagnostic performance of CadAI-B .Results: IDC revealed a significantly higher mean POM (0.365 0.306) than DCIS (0.126 0.196; P < 0.001).Features with significant POM differences included an irregular shape (IDC: 0.344 vs. DCIS: 0.135; P = 0.032), indistinct margins (0.318 vs. 0.012; P = 0.017), microlobulated margins (0.419 vs. 0.132; P = 0.049), hypoechoic pattern (0.345 vs. 0.135; P = 0.023), and parallel orientation (0.362 vs. 0.044; P = 0.021).IDC lesions were assigned more frequently to higher BI-RADS categories (4B, 4C, 5) with elevated POM outputs, whereas the DCIS cases retained a lower POM despite classification as BI-RADS 4A/4B. Conclusion: CadAI-B has the potential to assist in the real-time differentiation of IDC from DCIS by capturing feature-level and categorical differences in POM and BI-RADS outputs.This capability may support surgical planning and multidisciplinary decision-making by providing standardized risk estimates during scanning.Nevertheless, large-scale multicenter studies will be needed to confirm the clinical applicability.