The role of artificial intelligence in ultrasonographic diagnosis of liver cancer: Current status and future perspectives

Abstract

With the rapid advancement of artificial intelligence (AI) technologies, ultrasonography has undergone transformative progress in the diagnosis of liver cancer. This review summarizes the latest developments of AI-assisted ultrasonographic diagnosis of liver cancer, including commonly used ultrasound modalities (such as B-mode ultrasound and contrast-enhanced ultrasound) and their applicability across different patient populations. AI models have demonstrated superior performance in tasks such as distinguishing malignant from benign lesions, tumor subtyping, and multitask learning. They are particularly proficient in detecting small lesions, extracting quantitative imaging features, and minimizing subjective bias. Technological advancements in this field have evolved from traditional machine learning to deep learning and further to multimodal fusion approaches. The focus has shifted from static image analysis to dynamic video processing, from single-task models to multitask frameworks, and from model-centric development to clinical integration. Although many AI models have outperformed traditional diagnostic methods and even expert radiologists in performance, their clinical translation remains hindered by several challenges. These include the scarcity of high-quality annotated data, absence of standardized protocols, limited model interpretability, and complexities in multimodal data integration. Future directions should concentrate on establishing standardized multicenter databases, advancing privacy-preserving federated learning techniques, strengthening interdisciplinary collaboration, and conducting prospective clinical validation and real-world studies. As a non-invasive and efficient tool, AI-assisted ultrasonographic diagnosis of liver cancer holds great promise in the era of precision medicine.