[Paper Review] The Past, Current, and Future of Neonatal Intensive Care Units with Artificial Intelligence

Elif Keleş, Ulaş Bağcı|arXiv (Cornell University)|Feb 1, 2023

Neonatal Respiratory Health ResearchMedicine3 citations

TL;DR

This systematic review synthesizes 106 studies (1996–2022) on artificial intelligence in neonatal intensive care, evaluating classical machine learning and deep learning applications in neonatology. It analyzes AI in survival prediction, neuroimaging, vital sign monitoring, and retinopathy of prematurity, identifying key methodologies, challenges, and future integration roadmaps for NICUs using PRISMA 2020 guidelines.

ABSTRACT

Machine learning and deep learning are two subsets of artificial intelligence that involve teaching computers to learn and make decisions from any sort of data. Most recent developments in artificial intelligence are coming from deep learning, which has proven revolutionary in almost all fields, from computer vision to health sciences. The effects of deep learning in medicine have changed the conventional ways of clinical application significantly. Although some sub-fields of medicine, such as pediatrics, have been relatively slow in receiving the critical benefits of deep learning, related research in pediatrics has started to accumulate to a significant level, too. Hence, in this paper, we review recently developed machine learning and deep learning-based solutions for neonatology applications. We systematically evaluate the roles of both classical machine learning and deep learning in neonatology applications, define the methodologies, including algorithmic developments, and describe the remaining challenges in the assessment of neonatal diseases by using PRISMA 2020 guidelines. To date, the primary areas of focus in neonatology regarding AI applications have included survival analysis, neuroimaging, analysis of vital parameters and biosignals, and retinopathy of prematurity diagnosis. We have categorically summarized 106 research articles from 1996 to 2022 and discussed their pros and cons, respectively. In this systematic review, we aimed to further enhance the comprehensiveness of the study. We also discuss possible directions for new AI models and the future of neonatology with the rising power of AI, suggesting roadmaps for the integration of AI into neonatal intensive care units.

Motivation & Objective

• To comprehensively evaluate the role of machine learning and deep learning in neonatal intensive care units (NICUs) over the past three decades.
• To identify and categorize key AI applications in neonatology, including survival analysis, neuroimaging, biosignal processing, and retinopathy of prematurity (ROP) diagnosis.
• To assess methodological developments, algorithmic innovations, and remaining challenges in AI-driven neonatal disease assessment.
• To provide a structured roadmap for integrating AI into clinical NICU workflows using evidence-based review standards (PRISMA 2020).
• To project future directions for AI in neonatology, emphasizing clinical translation and system-level integration.

Proposed method

• Conducted a systematic literature review of 106 peer-reviewed studies on AI in neonatology from 1996 to 2022 using PRISMA 2020 guidelines.
• Categorized studies by application domain: survival analysis, neuroimaging, biosignal and vital parameter analysis, and ROP diagnosis.
• Evaluated algorithmic approaches, including classical machine learning (e.g., random forests, SVMs) and deep learning (e.g., CNNs, RNNs, Transformers) in clinical neonatal data.
• Synthesized methodological strengths and limitations, focusing on data quality, model interpretability, and clinical validation.
• Mapped the evolution of AI techniques in neonatal care, identifying trends and gaps in research and implementation.
• Proposed future integration roadmaps for AI in NICUs, emphasizing interoperability, real-time monitoring, and clinical decision support.

Experimental results

Research questions

• RQ1 What are the primary clinical applications of AI in neonatal intensive care units, and how have they evolved from 1996 to 2022?
• RQ2 How do classical machine learning and deep learning models compare in performance and clinical applicability for neonatal disease prediction and diagnosis?
• RQ3 What methodological challenges—such as data scarcity, model interpretability, and generalizability—limit the deployment of AI in neonatal care?
• RQ4 How can AI models be systematically integrated into NICU workflows to improve clinical outcomes and decision-making?
• RQ5 What future research directions and technical advancements are needed to enable scalable, real-world AI deployment in neonatology?

Key findings

• Deep learning has significantly advanced neonatal applications, particularly in neuroimaging and retinopathy of prematurity (ROP) detection, with some models achieving >90% accuracy in ROP screening.
• AI models for vital sign and biosignal analysis show promise in predicting sepsis and apnea, though clinical validation remains limited in large, diverse cohorts.
• Survival prediction models using machine learning demonstrate improved risk stratification over traditional scoring systems, especially when incorporating multimodal data.
• Despite progress, challenges in data quality, model interpretability, and lack of prospective clinical trials hinder widespread NICU adoption.
• The integration of AI into NICUs is still in early stages, with most studies remaining proof-of-concept; real-time, scalable deployment remains a key future challenge.
• Future AI development should prioritize explainable AI, federated learning for data privacy, and standardized benchmarks to ensure clinical reliability and regulatory compliance.