Artificial Intelligence in Neonatal Care: The Breadth of Promise, the Depth of Challenge-An Overview

Artificial intelligence (AI) is becoming an integral tool in clinical care. The recent position statement by the Royal Australasian College of Physicians (RACP) provides a timely practical blueprint on implementing and monitoring the use of AI in clinical practice. Although the data-rich environment of NICU presents a good setting for evaluation of AI-assisted clinical application, research on AI in Neonatology is relatively sparse compared to other fields in medicine. In this narrative overview, we summarised published research on AI in neonatology identified through a PubMed search to 29 December 2025, including early forms of AI/machine learning technologies in the 1990s to the more recent deep learning and large language model approaches across a broad range of conditions and functions. Our overview reveals a highly fragmented evidence landscape with the vast majority of published literature grouped around a few conditions, chiefly retinopathy of prematurity characterisation, neonatal seizure detection and neurodevelopmental outcome prediction, with very few studies in other areas. Most studies are single-centre, retrospective and preliminary, without any external validation. A small number of 'near-deployment' technologies, including automated oxygen control, the ANSeR seizure-detection algorithm, AI-based retinopathy of prematurity screening and automated General Movements Assessment for cerebral palsy screening, are notable exceptions. Rigorous clinical trials on AI-assisted therapy, diagnosis or prognostication, followed by implementation studies evaluating workflow integration across diverse geographical and socioeconomic settings are needed to address this critical translation-to-practice gap. Anticipated challenges specific to neonatology include methodological issues such as shifting physiological benchmarks across gestational ages and imperfect reference standards, as well as ethical issues such as establishing consent frameworks for long-term data use in a vulnerable population. Challenges shared with other fields in medicine include ensuring equitable applicability of AI across diverse populations with differing characteristics and resources and maintaining transparency in AI model development and deployment.