Artificial intelligence in assisted localization of fetal ultrasound standard planes: a narrative review
Background and Objective Ultrasound imaging has become a vital clinical diagnostic tool due to its cost-effectiveness and absence of ionizing radiation. In fetal ultrasound, the precise localization of standard ultrasound planes is a prerequisite for subsequent quantitative analysis and diagnosis. However, manual identification is highly operator-dependent and subjective, and the multi-planar data acquired by three-dimensional (3D) fetal ultrasound present challenges for efficient localization within large datasets. The objective of this narrative review is to systematically summarize the research progress and methodologies of artificial intelligence (AI) in the automated and precise localization of fetal ultrasound standard planes, while discussing the current challenges and future directions for its clinical application. Methods We conducted structured literature search in the PubMed and Web of Science databases to identify studies applying AI to ultrasound imaging, with a primary focus on fetal standard plane recognition. The search focused on all types of publications involving AI-based standard plane recognition in ultrasound between 1998 and 2025. Only peer-reviewed articles and review papers published in English were included. Titles and abstracts were screened to determine eligibility. Key Content and Findings This review details the transformative impact of AI, particularly deep learning, on automating standard plane localization in ultrasound, with a primary emphasis on fetal applications. Key findings reveal a progression from handcrafted features to advanced architectures like convolutional neural network and reinforcement learning, achieving expert-level accuracy across fetal standard planes, including abdominal, facial, brain, and cardiac views, while representative adult applications (e.g., hepatobiliary and thyroid imaging) are included as supplementary examples. The review identifies technical challenges, including data variability and computational costs, which are being addressed via transfer learning, attention mechanisms, and lightweight network design. The emergence of multi-organ models highlights a trend towards comprehensive AI systems for holistic screening. Conclusions AI significantly advances ultrasound by automating standard plane localization, particularly in fetal ultrasound, enhancing diagnostic consistency and workflow efficiency. Integration into clinical practice promises to reduce operator dependency and improve screening accessibility. Persistent challenges include model generalizability, data scarcity, and real-world clinical integration. Future work should prioritize multi-center validation, explainable AI, and end-to-end system development to fully realize this technology’s potential.