Use of Artificial Intelligence in Breast Ultrasound Imaging: Diagnosis and Clinical Decision Support

Breast ultrasound (US) is a critical non-invasive imaging modality for evaluating breast lesions, particularly in women with dense breast tissue. However, conventional interpretation suffers from inter-observer variability and high false-positive rates due to operator dependence and subjectivity. Objectives: To evaluate the role of Artificial Intelligence (AI), specifically deep learning models, in enhancing diagnostic accuracy, reducing unnecessary interventions, and supporting clinical decision-making in breast ultrasound imaging. Methods: A comprehensive review of recent literature (2000-2025) was conducted, focusing on AI applications in breast ultrasound for lesion detection, classification, segmentation, and clinical workflow integration. Results: AI systems, particularly convolutional neural networks, demonstrate diagnostic accuracy with area under the curve (AUC) values ranging from 0.92 to 0.98, often matching or exceeding expert radiologist performance. These systems achieve sensitivities and specificities typically exceeding 85%, with some studies reporting up to 98% sensitivity. AI integration reduces false-positive rates by up to 37% and unnecessary biopsies by approximately 28%. Beyond diagnosis, AI assists in lesion segmentation, BI-RADS classification consistency, and risk stratification. Portable AI-powered devices have shown promise in resource-limited settings, achieving 96-98% sensitivity. Integration of quantitative ultrasound parameters with AI enhances lesion differentiation and treatment planning. Conclusions: AI in breast ultrasound significantly improves diagnostic precision, workflow efficiency, and accessibility. Despite challenges including dataset diversity, model interpretability, and clinical integration, ongoing developments support AI as a valuable adjunct tool for enhancing breast cancer detection and supporting personalized patient management.