ARTIFICIAL INTELLIGENCE IN AUTOMATED BREAST ULTRASOUND: A CRITICAL REVIEW OF DIAGNOSTIC ACCURACY, CLINICAL UTILITY, AND CURRENT EVIDENCE

Abstract

Automated Breast Ultrasound (ABUS) is an established adjunctive breast-imaging technique, particularly relevant to women with dense breasts. Artificial intelligence (AI), including computer-aided detection and deep-learning approaches, is increasingly being investigated to improve lesion detection, characterization, reader consistency, and workflow. This critical review examines evidence published from January 2015 through August 2026, distinguishing studies that directly evaluate ABUS from broader breast-ultrasound AI studies. The evidence base demonstrates that ABUS can provide clinically useful supplemental detection, while AI-assisted interpretation may improve diagnostic discrimination and reduce reading time, particularly for less-experienced readers. However, the available evidence remains heterogeneous: studies differ in population, lesion- versus patient-level analysis, reference standards, breast-density distribution, AI role, and external validation. Importantly, several frequently cited AI studies concern handheld breast ultrasound or mammographic density classification rather than ABUS and therefore should not be treated as direct evidence for AI-ABUS performance. Current evidence supports AI-ABUS as a promising adjunct rather than a replacement for radiologist interpretation. Prospective, multicenter, multivendor, externally validated studies with transparent reporting and clinically relevant outcomes are required before routine implementation can be recommended.