Advances in Artificial Intelligence-Assisted Diagnosis (AI) in Ultrasound

Artificial intelligence-assisted diagnosis (AI) has emerged as a transformative tool in medical imaging, leveraging advanced computational techniques to enhance lesion detection, characterization, and differentiation between benign and malignant conditions. While AI has gained significant traction in CT and MRI, particularly for breast and pulmonary nodule diagnosis, its application in ultrasound remains relatively nascent. Ultrasound, despite its widespread clinical utility, is highly operator-dependent, with diagnostic accuracy influenced by subjective factors. AI-assisted systems aim to mitigate these limitations by providing objective, reproducible analyses. This review explores the fundamental principles of AI-assisted diagnostic systems and their evolving role in ultrasound imaging, with a focus on key clinical applications, challenges, and future directions. The review begins by outlining the historical development and workflow of AI-assisted diagnosis, emphasizing its three core steps: image preprocessing, feature extraction, and data processing. It then delves into specific applications across diverse ultrasound domains: Thyroid Nodules: AI systems demonstrate promise in automating malignancy risk stratification, though challenges persist in real-time performance and regional nodule identification. Breast Nodules: Integration with Breast Imaging Reporting and Data System (BI-RADS）criteria enables improved classification accuracy, particularly for less experienced sonographers. Liver Fibrosis: AI-driven texture analysis of liver parenchyma and capsule geometry offers non-invasive staging tools, albeit with technical complexities due to anatomical variability. Carotid Atherosclerosis: AI aids in plaque detection and stenosis quantification, reducing operator dependence in vascular ultrasound. Myocardial Infarction: Segmentation and acoustic feature analysis enhance echocardiographic assessment of infarcted myocardium, though standardized algorithms remain under development. Additional applications in musculoskeletal imaging, obstetrics, and endoscopic ultrasound are also highlighted. While AI-assisted ultrasound diagnosis holds immense potential to improve diagnostic accuracy, workflow efficiency, and early disease detection, its clinical adoption lags behind other imaging modalities due to challenges in image standardization and algorithm robustness. Future advancements hinge on large-scale validation studies, innovations in deep learning architectures, and interdisciplinary collaboration. With continued refinement, AI is poised to revolutionize ultrasound practice, bridging gaps in precision medicine and expanding access to high-quality diagnostic care.