Artificial Intelligence in Pediatric Healthcare - Part III: AI in Pediatric Inpatient and Critical Care

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Artificial intelligence (AI) is increasingly integrated into pediatric healthcare, supported by the expansion of electronic health records, continuous physiological monitoring, digital imaging, and computational methods capable of analyzing complex, time-dependent clinical data. This article is the third in a four-part series examining the evolving role of AI in pediatric medicine. The first article introduced the fundamental concepts, methodologies, data sources, and model development processes underlying AI in medicine, while the second reviewed its applications in ambulatory and preventive pediatrics. Building on these foundations, the present review examines AI applications in hospital-based pediatric care, with emphasis on pediatric emergency medicine, pediatric and neonatal intensive care, diagnostic imaging, pediatric subspecialties, and hospital workflow optimization. Current evidence indicates that machine learning and deep learning approaches may support emergency triage, early sepsis detection, prediction of clinical deterioration, physiological monitoring, ventilator management, diagnostic assessment, risk stratification, and treatment-response prediction. AI-based systems have also demonstrated potential applications in pediatric radiology, cardiology, oncology, and neurology. However, despite substantial technological progress, translation into routine clinical practice remains limited. Major challenges include small, heterogeneous pediatric datasets; age-dependent physiological variation; inadequate external validation; limited model interpretability; algorithmic bias; false-positive alerts; and difficulty integrating AI tools into established clinical workflows. Overall, AI has the potential to augment clinical decision-making and support the timely identification and management of critically ill children. However, prospective validation, external evaluation, transparency, and demonstration of clinically meaningful benefit are essential before widespread implementation.