An Emerging Role of Artificial Intelligence in Pediatric Neuroanesthesia

Jung-Bin Park; Young-Eun Jang

doi:10.3340/jkns.2026.0033

Journal of Korean Neurosurgical Society > Volume 69(3); 2026 > Article

Park and Jang: An Emerging Role of Artificial Intelligence in Pediatric Neuroanesthesia

Review Article

Pediatric

Journal of Korean Neurosurgical Society 2026; 69(3): 392-402.

Published online: April 9, 2026

DOI: https://doi.org/10.3340/jkns.2026.0033

An Emerging Role of Artificial Intelligence in Pediatric Neuroanesthesia

Jung-Bin Park

, Young-Eun Jang

Department of Anesthesiology and Pain Medicine, Seoul National University Hospital, Seoul National University College of Medicine, Seoul, Korea

Address for reprints : Young-Eun Jang Department of Anesthesiology and Pain Medicine, Seoul National University Hospital, Seoul National University College of Medicine, 101 Daehak-ro, Jongno-gu, Seoul 03080, Korea Tel : +82-2-2072-3650, Fax : +82-2-747-8364, E-mail : na0ag2@hotmail.com

Received: February 3, 2026 Revised: February 27, 2026 Accepted: April 4, 2026

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

Anesthesia for pediatric neurosurgery represents a highly complex and challenging field, characterized by age-dependent physiological variability, heterogeneous patient populations, and the critical need to protect the developing central nervous system. Conventional clinical approaches often rely on adult-derived data, which may inadequately reflect the distinct neurophysiological and hemodynamic characteristics of neonates, infants and children. Recent advances in artificial intelligence (AI) have enabled the integration of multimodal perioperative data, including physiologic signals and neurophysiologic monitoring, with the aim of supporting clinical decision-making in complex surgical settings. This review summarizes the current landscape of AI applications relevant to pediatric anesthesia, with particular attention to preoperative risk assessment, airway management, and real-time prediction of intraoperative adverse events such as hypoxemia and hemodynamic instability. Although AI-based approaches have demonstrated encouraging results in adult populations, their application to pediatric neuroanesthesia remains limited. The integration of AI into this field faces several distinct challenges, including the scarcity of high-quality datasets for rare neurosurgical conditions, substantial heterogeneity across developmental stages, and difficulties in aligning model outputs with clinically interpretable physiologic mechanisms. Addressing these limitations will require the development of explainable, physiology-informed AI frameworks and disease-specific models tailored to conditions such as moyamoya disease or complex craniofacial reconstruction. Ultimately, AI should be positioned as an adjunctive decision-support tool that complements, rather than replaces, anesthesiologists’ expertise. Through multidisciplinary collaboration and human-centered implementation, AI may contribute to improved perioperative safety and long-term neurodevelopmental outcomes in vulnerable pediatric neurosurgical patients.

Key Words: Artificial intelligence · Pediatric anesthesia · Perioperative care · Machine learning.

INTRODUCTION

Anesthesia for pediatric neurosurgical procedures represents one of the most challenging areas in modern anesthesiology [12,58,69]. Unlike adult practice, pediatric neuroanesthesia involves a highly heterogeneous patient population spanning from neonates to adolescents. This diversity is characterized not only by wide variations in anatomical scale, but also by age-dependent physiological profiles, evolving pharmacokinetics, and distinct risks of perioperative complications [12,58].

The primary clinical imperative in this field is the mitigation of risks associated with the heightened vulnerability of the developing central nervous system. Furthermore, the presence of multi-systemic comorbidities, such as those observed in patients with VACTERL association or syndromic craniosynostosis, often complicates airway management and hemodynamic stability [17,74]. Despite these challenges, current monitoring modalities and clinical protocols often rely on data extrapolated from adult populations or healthy children, which may fail to adequately account for the unique developmental trajectories and physiological variability of pediatric neurosurgical patients.

In this context, artificial intelligence (AI) has emerged as promising tools with the potential to augment clinical decisionmaking in high-stakes perioperative environments [14,26,47]. By enabling the integration and analysis of large volumes of multidimensional perioperative data, AI-based systems may help overcome limitations inherent to conventional rule-based approaches [26]. Recent studies in adult anesthesia have demonstrated the ability of AI-driven models to predict intraoperative hypotension and optimize anesthetic drug delivery; however, the application of these technologies within pediatric anesthesia—and particularly pediatric neuroanesthesia—remains at an early stage [27,33,42,48,72].

Despite these challenges, the convergence of AI and pediatric neuroscience offers a compelling opportunity to advance personalized and proactive anesthetic care. This review aims to explore the emerging role of AI in pediatric neuroanesthesia, with a focus on current applications in preoperative risk assessment and intraoperative monitoring. In addition, we discuss the unique methodological, clinical, and ethical challenges that must be addressed to enable the safe and effective implementation of AI-assisted decision-support systems in this vulnerable patient population.

UNIQUE ANESTHETIC CHALLENGES IN PEDIATRIC NEUROANESTHESIA : THE IMPETUS FOR AI INTEGRATION

Pediatric neuroanesthesia is uniquely complex, requiring precise anesthetic management to balance optimal surgical conditions with preservation of vulnerable neurophysiological function across a broad spectrum of developmental stages [10,12,41]. From neonates with transitional circulation to adolescents with near-adult physiology, age-dependent maturation of organ systems profoundly influences drug metabolism, cerebral autoregulation, and physiological reserve, necessitating highly individualized anesthetic strategies [10,41].

In many pediatric neurosurgical procedures, intraoperative neurophysiologic monitoring is commonly used to reduce the risk of neurological injury. In this context, total intravenous anesthesia is often preferred, as volatile anesthetics suppress neurophysiologic signals in a dose-dependent manner. However, the pharmacokinetics and pharmacodynamics (PK-PD) of intravenous anesthetic agents vary substantially with age, body composition, and organ maturation [4,6]. Consequently, achieving stable anesthetic depth while preserving reliable neurophysiologic monitoring remains technically demanding and highly operator dependent. Conventional anesthetic titration frequently relies on weight-based formulas or generalized nomograms, which inadequately capture the non-linear PK-PD relationships observed in infants and young children [3,5]. This challenge is further compounded by limitations of processed electroencephalography (EEG)-based monitoring. Delta-wave dominance in infants and the rapid evolution of EEG patterns across childhood render adult-calibrated algorithms less reliable, while interference from surgical manipulation and concurrent neurophysiologic monitoring further degrades signal interpretability [61,76,77]. Together, these factors complicate the construction of robust, physiology-based anesthetic depth models in pediatric neuroanesthesia.

Certain neurosurgical pathologies and procedures impose additional physiological constraints that further increase anesthetic complexity in pediatric neuroanesthesia, largely through their impact on cerebral autoregulation, intracranial dynamics, and oxygen delivery [10,41]. In children with moyamoya disease, meticulous maintenance of cerebral blood flow, oxygen delivery, and arterial carbon dioxide tension is critical, as impaired cerebrovascular reserve renders cerebral perfusion highly sensitive to even subtle hemodynamic perturbations, potentially precipitating catastrophic ischemic events [9,38,59]. Posterior fossa tumor surgery poses additional challenges, as patient positioning and direct brainstem manipulation may trigger abrupt autonomic disturbances, including severe bradycardia or sinus pauses [30]. In craniosynostosis surgery, anesthetic management is further complicated by a combination of elevated or vulnerable intracranial pressure, altered cranial compliance, and substantial intraoperative blood loss, all of which may compromise cerebral perfusion pressure if not carefully managed [67]. Across pediatric neurosurgical procedures, preservation of adequate hemoglobin concentration, cardiac output, cerebral blood flow, cerebral perfusion pressure, and arterial carbon dioxide tension is essential to prevent secondary neurological injury, particularly in the setting of immature or impaired cerebral autoregulation. Furthermore, these challenges are frequently compounded by the vulnerability of pediatric physiology. Infants and young children are especially susceptible to intraoperative hypoxemia caused by anesthesia-induced atelectasis, underscoring the importance of lung-protective ventilation and continuous respiratory monitoring, while prolonged prone positioning necessitates heightened vigilance to prevent pressure-related injuries [25,31,43,65]. Taken together, these scenarios generate complex, high-frequency, and interdependent physiologic data streams that must be interpreted continuously and in real time. Manual integration of such data is inherently vulnerable to cognitive overload, delayed recognition, and human error during critical intraoperative periods, highlighting the potential value of AI-assisted real-time risk detection and decision-support systems.

Pediatric neurosurgical patients frequently present with syndromic or rare conditions, such as VACTERL association or craniofacial anomalies, introducing multisystem complexities including difficult airways, congenital cardiac anomalies, and altered intracranial compliance [51,53]. In many of these rare disorders, anesthetic management is informed not by standardized guidelines but by fragmented evidence derived from case reports, small case series, and individual institutional experience. As a result, pediatric anesthesiologists are required to integrate a wide range of heterogeneous and condition-specific considerations, including airway anatomy, cardiovascular physiology, cerebral autoregulation, and intracranial dynamics. The safe management of these complex and interacting factors relies heavily on substantial clinical experience and expertise, particularly during critical intraoperative periods or when unexpected physiological changes arise [24].

Taken together, pediatric neuroanesthesia generates dynamic, high-dimensional data encompassing hemodynamics, ventilation, neurophysiology, and anesthetic drug delivery. The complexity and temporal variability of these data often exceed the limits of conventional decision-making and real-time human cognition, particularly during critical intraoperative periods. In this context, AI-based approaches may offer a complementary means of integrating multimodal physiologic information, supporting individualized anesthetic management, and facilitating earlier recognition of physiological instability, thereby enhancing perioperative safety in pediatric neuroanesthesia.

CURRENT STATE OF AI IN PEDIATRIC ANESTHESIA

To date, the clinical translation of AI has advanced more rapidly in adult anesthesiology, where several well-established systems have progressed beyond experimental validation. Among the most widely recognized examples is the arterial waveform-based Hypotension prediction index, a commercially available system that utilizes machine learning (ML) algorithms to forecast impending intraoperative hypotension before clinically apparent blood pressure decline [71]. In addition, large-scale multimodal ML models such as MySurgeryRisk and other perioperative complication prediction systems have been developed using electronic health record and intraoperative physiological data to predict postoperative morbidity, acute kidney injury, respiratory failure, and intensive care unit admission [11]. These models have shown improved predictive performance compared with traditional risk calculators and have influenced perioperative risk stratification strategies. Collectively, these established adult models demonstrate the progression of AI in anesthesia from physiologic prediction to automated control and perioperative risk analytics, providing a translational roadmap for future development in pediatric neuroanesthesia.

Building upon these established foundations, the integration of AI into pediatric anesthesia has thus far focused primarily on enhancing perioperative safety and clinical efficiency through predictive modeling and decision-support systems (Fig. 1). Although the pediatric literature remains less extensive than that in adult anesthesia, accumulating evidence suggests that AI-based tools may be particularly valuable in pediatric populations, where physiological variability, limited physiological reserve, and narrow margins for error increase the risk of perioperative complications [48].

Preoperative risk stratification

ML models using electronic medical record (EMR) data have been developed to identify children at increased risk of perioperative respiratory, cardiovascular, and neurological complications. Using the large, multicenter APRICOT dataset, investigators developed ML models to identify children classified as American Society of Anesthesiologists (ASA) physical status IIII who were at low risk for severe perioperative critical events to guide decisions regarding the appropriate level of anesthetic supervision and perioperative care [23]. In addition, ML has enabled the development of pediatric risk assessment scores for predicting perioperative morbidity and mortality [52,54,55].

Airway management

Airway management represents another domain in which AI has shown substantial promise. Pediatric airways differ markedly from adult airways in both anatomy and physiology, and inaccurate endotracheal tube (ETT) sizing or insertion depth can lead to hypoxemia, airway trauma, or inadequate ventilation. ML models incorporating patient demographics, anthropometric measurements, and airway-related variables have demonstrated improved accuracy compared with traditional formula-based methods for predicting optimal ETT size and depth [36,79].

More recently, ML-based approaches have been developed to identify and label vocal cords and tracheal anatomy in real time using laryngoscopic or bronchoscopic imaging [37,49]. These systems may provide guidance during laryngoscopy to assist novice clinicians managing difficult pediatric airways.

In addition, computer vision-based algorithms using facial morphology and image-based analysis have been explored to identify difficult airways more accurately than conventional clinical assessment [15,16]. These approaches are particularly relevant in pediatric patients with craniofacial abnormalities or syndromic conditions, which are frequently encountered in pediatric neurosurgical practice.

Perioperative respiratory adverse event

Perioperative respiratory adverse events, including upper airway obstruction and hypoxemia, represent a major source of morbidity in pediatric anesthesia. Recent studies have explored the use of neural network-based model using photoplethysmography waveform data to identify postoperative airway obstruction, demonstrating the feasibility of non-invasive, signalbased detection of upper airway compromise [39]. Sippl et al. [62] developed ML models to predict postintubation hypoxic episodes during general anesthesia, enabling early identification of patients at increased risk for respiratory deterioration. Also, several studies have investigated AI-based models for the classification and prediction of perioperative respiratory adverse events in pediatric populations, highlighting the potential of data-driven approaches to support early intervention and risk mitigation [66,78].

Intraoperative hypoxemia represents a time-critical clinical event requiring prompt intervention to prevent adverse outcomes. Pediatric patients are particularly vulnerable to hypoxemia due to limited physiological reserves, including reduced functional residual capacity and increased metabolic demand with higher oxygen consumption [18]. These characteristics contribute to more rapid desaturation and delayed recovery compared with adults. Consequently, real-time prediction of hypoxemic events during general anesthesia remains challenging. Park et al. [56] demonstrated that Gradient-Boosting Machine based ML models can predict imminent intraoperative hypoxemia using high-resolution biosignals obtained from standard patient monitors, enabling prediction approximately one minute in advance. Notably, the predictive performance of these models was confirmed through external validation, supporting their potential utility in clinical practice [8].

Hemodynamic monitoring

Hemodynamic monitoring and prediction constitute another important area of AI application in anesthesiology. Maintenance of adequate cardiac output and cerebral perfusion is critical during pediatric neurosurgery, yet continuous and reliable measurement remains challenging. Several AI-based models using arterial waveform analysis have been developed to estimate cardiac output and predict impending hypotension, primarily in adult surgical and critical care populations [27,33,64]. To date, however, pediatric surgical population-specific AI models for real-time hemodynamic estimation and prediction remain scarce. Despite promising results in adult studies, translation of these approaches to pediatric neuroanesthesia has been limited, highlighting a critical gap in current AI research.

Transfusion risk prediction

Blood transfusion remains a significant concern in pediatric surgical procedures associated with substantial blood loss, particularly in younger patients with limited physiological reserve. Recent studies have explored the predictive models to stratify perioperative blood transfusion risk with clinically meaningful accuracy in children undergoing surgery for developmental dysplasia of the hip [2].

Monitoring of anesthetic depth and drug administration

Assessment of anesthetic depth and brain state has also been explored using AI. Conventional EEG-derived indices, such as the bispectral index (Medtronic, Dublin, Ireland) and patient state index (Masimo Corp., Irvine, CA, USA), have recognized limitations in children due to age-dependent EEG characteristics and interference from surgical and neurophysiologic monitoring [75,77]. Two studies reported the use of supervised ML classifiers to classify anesthetic states using EEG signals, primarily focused on discriminating between awake and anesthetized states rather than generating a continuous depth index [13,35].

Beyond predictive monitoring, closed-loop anesthetic delivery systems incorporating AI algorithms have been developed to maintain stable anesthetic depth through automated drug titration [19,28,46,68,70,72]. To date, however, most clinical implementations and validation studies have been conducted in adult populations.

Postoperative outcome prediction using intraoperative variables

AI-based models that integrate multimodal perioperative data—including intraoperative vital signs, laboratory values, and intraoperative events—have demonstrated the ability to improve prediction of postoperative outcomes and adverse events [1,63,73]. Such predictive capability may support proactive postoperative management strategies aimed at reducing morbidity. Recent work has demonstrated the feasibility of explainable ML approaches for postoperative outcome prediction in neonatal surgical patients [29]. In this study, intraoperative vital signs emerged as major contributors to postoperative survival, highlighting the critical role of dynamic perioperative physiology in outcome prediction [29]. In addition, tree-based ML approaches have been used to identify intraoperative hemodynamic targets associated with early postoperative organ function in pediatric patients, such as weight-adjusted blood pressure thresholds linked to improved graft function after kidney transplantation [50].

Operating room management

Efficient operating room management is a critical component of pediatric anesthesia, as delays and last-minute surgical cancellations are associated with increased healthcare costs, fasting time as well as significant psychological stress for children and their caregivers [40,57]. Recent studies have demonstrated that ML model capable of predicting last-minute cancellation of elective pediatric surgeries with acceptable accuracy [44,45]. In addition, ML-based forecasting of surgical duration has emerged as a promising strategy to optimize operating room scheduling and resource utilization, with the potential to substantially reduce delays [21,32,57]. Park et al. [57] demonstrated that explainable AI-driven prediction models incorporating detailed perioperative and provider-related variables achieved superior performance compared with conventional scheduling methods. These findings suggest that context-aware model development can markedly enhance prediction accuracy.

CHALLENGES AND BARRIERS TO AI IMPLEMENTATION IN PEDIATRIC NEUROANESTHESIA

Despite the growing interest in AI for perioperative decision support, several substantial barriers currently limit its effective implementation in pediatric neurosurgical anesthesia. These challenges reflect both the intrinsic characteristics of pediatric neurosurgical populations and broader limitations of AI development in pediatric anesthesia

A fundamental challenge lies in the availability of sufficiently large and high-quality datasets [48]. Pediatric neurosurgical conditions—including congenital malformations, rare and diverse tumor subtypes, and complex cerebrovascular diseases—are relatively uncommon, making it difficult for single institutions to accumulate datasets of adequate size for robust AI model training and validation. As a result, many existing AI models are underpowered, lack external validation, and may not generalize across institutions or patient populations [7]. The predominance of single-center studies and small sample sizes further exacerbates this “small data” problem, limiting not only model scalability but also the rigorous assessment of external validity, feasibility, and real-world clinical efficacy [7]. Consequently, demonstrating reliable performance across diverse clinical settings and establishing meaningful clinical benefit remain substantial challenges for AI applications in pediatric neuroanesthesia.

Clinical heterogeneity further complicates AI implementation. Pediatric neurosurgical patients frequently present congenital syndromes or multisystem comorbidities that influence anesthetic risk through complex interactions between neurological, respiratory, cardiovascular, and airway factors. Moreover, rapid physiological changes from infancy through adolescence reduce the generalizability of models trained on narrow age ranges, leading to fragmented datasets that hinder the development of universal AI tools.

In addition, perioperative monitoring practices in pediatric neuroanesthesia are not standardized across institutions. Variability in anesthetic techniques and monitoring configurations—including EEG-based depth-of-anesthesia monitoring, cerebral oximetry, invasive catheters, cardiac output estimation, and intraoperative neurophysiologic monitoring—introduces substantial inconsistency in data acquisition. Such heterogeneity in monitoring modalities and signal quality complicates data harmonization and poses a significant barrier to the development of generalizable AI models.

Another major limitation is the lack of longitudinal outcome data. In pediatric neuroanesthesia, clinically meaningful outcomes often extend beyond the immediate perioperative period and include long-term neurodevelopmental, cognitive, and functional recovery. However, standardized longitudinal follow-up is rarely available, and most datasets rely on short-term or surrogate endpoints. This disconnect limits the ability of AI models to capture outcomes that are most relevant to the developing brain and undermines their clinical applicability.

An additional and often underappreciated challenge is the difficulty of defining reliable ground truth labels. Fundamental concepts such as “cerebral ischemia,” “adequate anesthesia,” or “neurological injury” lack universally accepted operational definitions in the perioperative setting. Outcomes are frequently inferred from indirect markers or transient physiological changes rather than definitive neurological endpoints. This variability in outcome definitions introduces noise into model training, reduces reproducibility across studies, and challenges the clinical validity of AI predictions.

In a high-stakes environment like the operating room, pediatric anesthesiologists must understand the reasoning behind a clinical recommendation. However, many advanced high-performing AI algorithms, particularly deep learning models, operate as "black boxes”, where the process of reaching a conclusion is not transparent to human users. For AI to be integrated into pediatric neuroanesthesia, explainable AI is essential [20]. Clinicians must be able to contextualize AI recommendations within established physiological principles and individual patient characteristics to ensure patient safety.

Ethical and regulatory considerations also warrant careful attention [34]. Pediatric AI research raises heightened concerns regarding data privacy, consent, and algorithmic bias, particularly for rare diseases and underrepresented populations [22,34,60]. Models trained on limited or unbalanced datasets may inadvertently reinforce existing disparities or produce unreliable predictions when applied outside their development context. Data sharing across institutions, which is necessary to overcome data scarcity, often faces strict privacy regulations. Additionally, there is a risk of algorithmic bias, where models trained on biased or unrepresentative data may provide inaccurate recommendations for certain ethnic or age groups, potentially compromising the safety of pediatric patients.

In conclusion, these barriers highlight that successful AI integration in pediatric neuroanesthesia will require more than algorithmic innovation alone. Progress will depend on collaborative efforts to establish multicenter pediatric datasets, standardize perioperative data collection, and develop AI models that are transparent, interpretable, and explicitly tailored to pediatric neurophysiology and anesthetic practice.

FUTURE DIRECTIONS OF AI RESEARCH IN

PEDIATRIC NEUROSURGICAL ANESTHESIA The ultimate goal of AI in pediatric neuroanesthesia is to enhance perioperative neurological safety while supporting optimal neurodevelopmental outcomes in a highly vulnerable population. Given the narrow physiological margins and dynamic intraoperative conditions characteristic of pediatric neurosurgery, future AI research should focus on clinically interpretable, physiology-informed decision-support systems that complement pediatric anesthesiologists’ expertise rather than replace clinical judgment. In this context, the development of explainable AI frameworks is essential to ensure that AI-driven recommendations remain transparent, clinically interpretable, and aligned with established principles of pediatric neurophysiology.

Despite the substantial challenges outlined above, the potential directions in which AI may contribute to pediatric neuroanesthesia are increasingly clear (Fig. 2). One promising avenue is the development of disease-specific AI models tailored to the unique pathophysiology and perioperative risks of individual neurosurgical conditions. For example, in pediatric patients with moyamoya disease, AI-driven systems may support individualized cerebral oxygen delivery management by continuously integrating hemoglobin concentration, cardiac output, blood pressure, and arterial carbon dioxide tension. Such models could assist clinicians in anticipating ischemic risk and proactively adjusting anesthetic, hemodynamic, and transfusion strategies based on each patient’s physiological reserve and disease severity.

Similarly, in pediatric brain tumor surgery, AI-based predictive models may assist anesthetic planning, and predict bleeding risk and cardiovascular instability. By incorporating tumor characteristics, anatomical relationships with major vessels, and surgical data, these systems could help anesthesiologists optimize fluid and transfusion strategies and maintain stable cerebral perfusion during prolonged procedures with less intraoperative adverse events.

In craniosynostosis surgery, where difficult airway management and significant blood loss are common, disease-specific AI tools may assist in perioperative airway management risk stratification and intraoperative transfusion planning, thereby supporting proactive anesthetic decision-making.

In addition, future AI research should prioritize the development of multimodal, real-time decision-support systems. Pediatric neuroanesthesia generates complex streams of physiological and clinical data, including EEG signals, hemodynamic variables, laboratory values, and intraoperative neurophysiologic monitoring. AI models capable of integrating these multimodal inputs in real time may function as a real time assistant, enhancing situational awareness and supporting intraoperative decision-making during critical surgical phases. Importantly, such systems should be designed to augment clinician judgment by presenting interpretable risk estimates and actionable insights.

In parallel with these conceptual directions, our institution has initiated several exploratory efforts aimed at gradually integrating AI into pediatric neuroanesthesia workflows. As part of this initiative, we developed a ML-based operating room scheduling model incorporating our institutional workflow variables and surgeon-specific factors. The model was designed to improve prediction of anesthesia start times and surgical duration, thereby facilitating reduction of unnecessary preoperative fasting time in children undergoing general anesthesia. We have also developed an AI-assisted preoperative evaluation support system that utilizes EMR data to generate structured preliminary assessments prior to surgery, and preparations are underway for integration into our institutional EMR platform. Beyond traditional ASA physical status classification, we are developing an AI-based risk stratification framework that incorporates patient-specific comorbidities, procedure-related factors, and current clinical conditions to estimate intraoperative risk profiles. Such a system is intended to support tailored anesthetic preparation, optimized resource allocation, equipment planning, personnel deployment, and anticipatory postoperative care. In the intraoperative setting, arterial waveform-based hypotension prediction systems are being assessed to facilitate anticipatory hemodynamic management. Exploratory analyses of high-resolution arterial line waveforms in pediatric patients with moyamoya disease are also underway to identify subtle hemodynamic variability patterns associated with postoperative neurological complications. They have contributed to heightened vigilance, closer physiologic monitoring, and more proactive adjustment of ventilation and vasoactive support during critical surgical phases. Despite legal, regulatory, and practical constraints that currently limit full-scale integration of AI systems into existing EMR platforms and clinical devices, these initiatives underscore the promising potential of incremental and responsibly implemented AI applications in pediatric neuroanesthesia.

In summary, AI is unlikely to replace clinical expertise in pediatric neuroanesthesia. Rather, through cautious and stepwise implementation, AI has the potential to evolve into a supportive tool that integrates complex clinical information, enhances patient safety, and ultimately contributes to improved anesthetic management and long-term neurological outcomes in children undergoing neurosurgical procedures.

CONCLUSION

Pediatric neuroanesthesia is uniquely demanding, requiring meticulous physiologic control to protect the developing brain while accommodating complex surgical and neurophysiologic constraints. Although AI has demonstrated growing potential across pediatric anesthesia and neurosurgery, its application in pediatric neuroanesthesia remains at an early stage, constrained by limited data availability, heterogeneity of disease, and challenges in model interpretability.

Progress in this field will depend on close multidisciplinary collaboration, pediatric-specific dataset development, and careful alignment of AI tools with established principles of neuroprotection and anesthetic practice. With a cautious and stepwise approach, AI has the potential to support safer, more individualized anesthetic care and ultimately improve neurological and developmental outcomes for children undergoing neurosurgical procedures. Rather than pursuing fully autonomous systems, early efforts should focus on targeted, physiology-informed decision-support tools that integrate surgical, anesthetic, and multimodal physiologic data to enhance situational awareness and early warning. By bridging the gap between advanced technology and clinical expertise, AI has the potential to significantly improve both the perioperative safety and the long-term developmental outcomes of pediatric neurosurgical patients.

Notes

Conflicts of interest

No potential conflict of interest relevant to this article was reported.

Informed consent

This type of study does not require informed consent.

Author contributions

Conceptualization : YEJ, JBP; Visualization : YEJ, JBP; Writing - original draft : YEJ, JBP; Writing - review & editing : YEJ, JBP

Data sharing

None

Preprint

None

Fig. 1.

Overview of an explainable artificial intelligence framework for pediatric neuroanesthesia. Multimodal perioperative data, including physiologic signals, electroencephalography, laboratory values, electronic medical records, and imaging data, undergo preprocessing steps such as labeling, artifact removal, segmentation, normalization, windowing, and epoching. The preprocessed data are used for model development with training, hyperparameter tuning, and internal validation. Post-hoc model interpretation and feature attribution are performed to enhance explainability and interpretability of model predictions. The finalized model is subsequently evaluated in an independent external validation cohort to assess generalizability and transportability, followed by application to real-world clinical settings. Clinical implementation focuses on decision support under human supervision, with downstream evaluation of clinically meaningful outcomes. EEG : electroencephalography, EMR : electronic medical record.

Fig. 2.

Schematic overview of potential applications of artificial intelligence across the perioperative continuum in pediatric neuroanesthesia. ICU : intensive care unit, EEG : electroencephalography.

References

1. Acosta JN, Falcone GJ, Rajpurkar P, Topol EJ : Multimodal biomedical AI. Nat Med 28 : 1773-1784, 2022

2. Alassaf N, Reitsma JB : Development of a prediction model for allogenic blood transfusion in children undergoing surgery for developmental dysplasia of the hip. Technol Health Care 27 : 327-334, 2019

3. Anderson BJ : Pediatric models for adult target-controlled infusion pumps. Paediatr Anaesth 20 : 223-232, 2010

4. Anderson BJ : Pharmacokinetics and pharmacodynamics in the pediatric population in Mason KP (eds) : Mason’s Pediatric Sedation Outside of the Operating Room, Cham : Springer, 2025, pp225-256

5. Anderson BJ, Bagshaw O : Practicalities of total intravenous anesthesia and target-controlled infusion in children. Anesthesiology 131 : 164-185, 2019

6. Anderson BJ, Morse JD, Hannam JA, Cortinez LI : Pharmacokinetic and pharmacodynamic considerations of general anesthesia in pediatric subjects. Expert Opin Drug Metab Toxicol 16 : 279-295, 2020

7. Antel R, Sahlas E, Gore G, Ingelmo P : Use of artificial intelligence in paediatric anaesthesia: a systematic review. BJA Open 5 : 100125, 2023

8. Baek S, Park JB, Heo J, Kim K, Baek D, Oh C, et al : Hypoxemia prediction in pediatric patients under general anesthesia using machine learning: a retrospective observational study and external validation. PLoS One 21 : e0339276, 2026

9. Bang OY, Fujimura M, Kim SK : The pathophysiology of moyamoya disease: an update. J Stroke 18 : 12-20, 2016

10. Benzon HA, Butler CG, Soriano SG : Advances in pediatric neuroanesthesia practices. Best Pract Res Clin Anaesthesiol 38 : 127-134, 2024

11. Bihorac A, Ozrazgat-Baslanti T, Ebadi A, Motaei A, Madkour M, Pardalos PM, et al : MySurgeryRisk: development and validation of a machine-learning risk algorithm for major complications and death after surgery. Ann Surg 269 : 652-662, 2019

12. Bissonnette B, Brady KM, Easley RB : Anesthesia for neurosurgical procedures in Gregory GA, Andropoulos DB (eds) : Gregory’s Pediatric Anesthesia, Fifth Edition, West Sussex : John Wiley & Sons Ltd, 2012, pp540-568

13. Chini M, Gretenkord S, Kostka JK, Pöpplau JA, Cornelissen L, Berde CB, et al : Neural correlates of anesthesia in newborn mice and humans. Front Neural Circuits 13 : 38, 2019

14. Connor CW : Artificial intelligence and machine learning in anesthesiology. Anesthesiology 131 : 1346-1359, 2019

15. Connor CW, Segal S : Accurate classification of difficult intubation by computerized facial analysis. Anesth Analg 112 : 84-93, 2011

16. Cuendet GL, Schoettker P, Yüce A, Sorci M, Gao H, Perruchoud C, et al : Facial image analysis for fully automatic prediction of difficult endotracheal intubation. IEEE Trans Biomed Eng 63 : 328-339, 2016

17. Cuperus IE, Bernard SE, Joosten KFM, Wolvius EB, Pullens B, Mathijssen IMJ, et al : Perioperative airway management for midface surgery in children with syndromic craniosynostosis; a single center experience with immediate extubation. Paediatr Anaesth 35 : 380-388, 2025

18. de Graaff JC, Bijker JB, Kappen TH, van Wolfswinkel L, Zuithoff NP, Kalkman CJ : Incidence of intraoperative hypoxemia in children in relation to age. Anesth Analg 117 : 169-175, 2013

19. De Smet T, Struys MM, Neckebroek MM, Van den Hauwe K, Bonte S, Mortier EP : The accuracy and clinical feasibility of a new bayesian-based closed-loop control system for propofol administration using the bispectral index as a controlled variable. Anesth Analg 107 : 1200-1210, 2008

20. Dwivedi R, Dave D, Naik H, Singhal S, Omer R, Patel P, et al : Explainable AI (XAI): core ideas, techniques, and solutions. ACM Comput Surv 55 : 1-33, 2023

21. Fairley M, Scheinker D, Brandeau ML : Improving the efficiency of the operating room environment with an optimization and machine learning model. Health Care Manag Sci 22 : 756-767, 2019

22. Farhud DD, Zokaei S : Ethical issues of artificial intelligence in medicine and healthcare. Iran J Public Health 50 : i-v, 2021

23. Gray GM, Ahumada LM, Rehman MA, Varughese A, Fernandez AM, Fackler J, et al : A machine-learning approach for decision support and risk stratification of pediatric perioperative patients based on the APRICOT dataset. Paediatr Anaesth 33 : 710-719, 2023

24. Habre W, Disma N, Virag K, Becke K, Hansen TG, Jöhr M, et al : Incidence of severe critical events in paediatric anaesthesia (APRICOT): a prospective multicentre observational study in 261 hospitals in Europe. Lancet Respir Med 5 : 412-425, 2017

25. Haisley M, Sørensen JA, Sollie M : Postoperative pressure injuries in adults having surgery under general anaesthesia: systematic review of perioperative risk factors. Br J Surg 107 : 338-347, 2020

26. Hashimoto DA, Witkowski E, Gao L, Meireles O, Rosman G : Artificial intelligence in anesthesiology: current techniques, clinical applications, and limitations. Anesthesiology 132 : 379-394, 2020

27. Hatib F, Jian Z, Buddi S, Lee C, Settels J, Sibert K, et al : Machine-learning algorithm to predict hypotension based on high-fidelity arterial pressure waveform analysis. Anesthesiology 129 : 663-674, 2018

28. Hemmerling TM, Arbeid E, Wehbe M, Cyr S, Taddei R, Zaouter C : Evaluation of a novel closed-loop total intravenous anaesthesia drug delivery system: a randomized controlled trial. Br J Anaesth 110 : 1031-1039, 2013

29. Hu Y, Gong X, Shu L, Zeng X, Duan H, Luo Q, et al : Understanding risk factors for postoperative mortality in neonates based on explainable machine learning technology. J Pediatr Surg 56 : 2165-2171, 2021

30. Jagannathan S, Krovvidi H : Anaesthetic considerations for posterior fossa surgery. CEACCP 14 : 202-206, 2014

31. Jang YE, Ji SH, Kim EH, Lee JH, Kim JT, Kim HS : Effect of regular alveolar recruitment on intraoperative atelectasis in paediatric patients ventilated in the prone position: a randomised controlled trial. Br J Anaesth 124 : 648-655, 2020

32. Jiao Y, Sharma A, Ben Abdallah A, Maddox TM, Kannampallil T : Probabilistic forecasting of surgical case duration using machine learning: model development and validation. J Am Med Inform Assoc 27 : 1885-1893, 2020

33. Kendale S, Kulkarni P, Rosenberg AD, Wang J : Supervised machine-learning predictive analytics for prediction of postinduction hypotension. Anesthesiology 129 : 675-688, 2018

34. Keskinbora KH : Medical ethics considerations on artificial intelligence. J Clin Neurosci 64 : 277-282, 2019

35. Khan FH, Saadeh W : An EEG-based hypnotic state monitor for patients during general anesthesia. IEEE Trans Very Large Scale Integr Syst 29 : 950-961, 2021

36. Kim H, Yoon HK, Lee H, Jung CW, Lee HC : Predicting optimal endotracheal tube size and depth in pediatric patients using demographic data and machine learning techniques. Korean J Anesthesiol 76 : 540-549, 2023

37. Kim JS, Cho Y, Lim TH : Prediction of the location of the glottis in laryngeal images by using a novel deep-learning algorithm. IEEE Access 7 : 79545-79554, 2019

38. Kim SK, Cho BK, Phi JH, Lee JY, Chae JH, Kim KJ, et al : Pediatric moyamoya disease: an analysis of 410 consecutive cases. Ann Neurol 68 : 92-101, 2010

39. Knorr-Chung BR, McGrath SP, Blike GT : Identifying airway obstructions using photoplethysmography (PPG). J Clin Monit Comput 22 : 95-101, 2008

40. Koh WX, Phelan R, Hopman WM, Engen D : Cancellation of elective surgery: rates, reasons and effect on patient satisfaction. Can J Surg 64 : E155-E161, 2021

41. Lamsal R, Rath GP : Pediatric neuroanesthesia. Curr Opin Anaesthesiol 31 : 539-543, 2018

42. Lee HC, Ryu HG, Chung EJ, Jung CW : Prediction of bispectral index during target-controlled infusion of propofol and remifentanil: a deep learning approach. Anesthesiology 128 : 492-501, 2018

43. Lee JH, Choi S, Ji SH, Jang YE, Kim EH, Kim HS, et al : Effect of an ultrasound-guided lung recruitment manoeuvre on postoperative atelectasis in children: a randomised controlled trial. Eur J Anaesthesiol 37 : 719-727, 2020

44. Li C, Li Z, Huang S, Chen X, Zhang T, Zhu J : Machine learning-based approach to predict last-minute cancellation of pediatric day surgeries. Comput Inform Nurs 42 : 363-368, 2024

45. Liu L, Ni Y, Zhang N, Nick Pratap J : Mining patient-specific and contextual data with machine learning technologies to predict cancellation of children’s surgery. Int J Med Inform 129 : 234-241, 2019

46. Liu Y, Li M, Yang D, Zhang X, Wu A, Yao S, et al : Closed-loop control better than open-loop control of profofol TCI guided by BIS: a randomized, controlled, multicenter clinical trial to evaluate the CONCERT-CL closed-loop system. PLoS One 10 : e0123862, 2015

47. Lopes S, Rocha G, Guimarães-Pereira L : Artificial intelligence and its clinical application in anesthesiology: a systematic review. J Clin Monit Comput 38 : 247-259, 2024

48. Madadi F, Dabbagh A, Sabouri AS : Artificial intelligence in pediatric anesthesia. Anesthesiol Clin 43 : 453-469, 2025

49. Matava C, Pankiv E, Raisbeck S, Caldeira M, Alam F : A convolutional neural network for real time classification, identification, and labelling of vocal cord and tracheal using laryngoscopy and bronchoscopy video. J Med Syst 44 : 44, 2020

50. Michelet D, Brasher C, Marsac L, Zanoun N, Assefi M, Elghoneimi A, et al : Intraoperative hemodynamic factors predicting early postoperative renal function in pediatric kidney transplantation. Paediatr Anaesth 27 : 927-934, 2017

51. Nargozian C : The airway in patients with craniofacial abnormalities. Paediatr Anaesth 14 : 53-59, 2004

52. Nasr VG, DiNardo JA, Faraoni D : Development of a pediatric risk assessment score to predict perioperative mortality in children undergoing noncardiac surgery. Anesth Analg 124 : 1514-1519, 2017

53. Nasr VG, Markham LW, Clay M, DiNardo JA, Faraoni D, Gottlieb-Sen D, et al : Perioperative considerations for pediatric patients with congenital heart disease presenting for noncardiac procedures: a scientific statement from the American Heart Association. Circ Cardiovasc Qual Outcomes 16 : e000113, 2023

54. Nasr VG, Staffa SJ, Zurakowski D, DiNardo JA, Faraoni D : Pediatric risk stratification is improved by integrating both patient comorbidities and intrinsic surgical risk. Anesthesiology 130 : 971-980, 2019

55. Nasr VG, Valencia E, Staffa SJ, Faraoni D, DiNardo JA, Berry JG, et al : Comprehensive risk assessment of morbidity in pediatric patients undergoing noncardiac surgery: an institutional experience. Anesth Analg 131 : 1607-1615, 2020

56. Park JB, Lee HJ, Yang HL, Kim EH, Lee HC, Jung CW, et al : Machine learning-based prediction of intraoperative hypoxemia for pediatric patients. PLoS One 18 : e0282303, 2023

57. Park JB, Roh GH, Kim K, Kim HS : Development of predictive model of surgical case durations using machine learning approach. J Med Syst 49 : 8, 2025

58. Paterson N, Waterhouse P : Risk in pediatric anesthesia. Paediatr Anaesth 21 : 848-857, 2011

59. Phi JH, Wang KC, Lee JY, Kim SK : Moyamoya syndrome: a window of moyamoya disease. J Korean Neurosurg Soc 57 : 408-414, 2015

60. Raheja A, Belani KG : The practical and ethical implications of artificial intelligence in anesthesiology, pain medicine, and intensive care: safeguards. Anesthesiol Clin 43 : 613-623, 2025

61. Ricci Z, Robino C, Rufini P, Cumbo S, Cavallini S, Gobbi L, et al : Monitoring anesthesia depth with patient state index during pediatric surgery. Paediatr Anaesth 33 : 855-861, 2023

62. Sippl P, Ganslandt T, Prokosch HU, Muenster T, Toddenroth D : Machine learning models of post-intubation hypoxia during general anesthesia in Röhrig R, Timmer A, Binder H, Sax U (eds) : German Medical Data Sciences: Visions and Bridges, Amsterdam : IOS Press, 2017, pp212-216

63. Soenksen LR, Ma Y, Zeng C, Boussioux L, Villalobos Carballo K, Na L, et al : Integrated multimodal artificial intelligence framework for healthcare applications. NPJ Digit Med 5 : 149, 2022

64. Solomon SC, Saxena RC, Neradilek MB, Hau V, Fong CT, Lang JD, et al : Forecasting a crisis: machine-learning models predict occurrence of intraoperative bradycardia associated with hypotension. Anesth Analg 130 : 1201-1210, 2020

65. Song IK, Jang YE, Lee JH, Kim EH, Yoo S, Kim HS, et al : Effect of different fraction of inspired oxygen on development of atelectasis in mechanically ventilated children: a randomized controlled trial. Paediatr Anaesth 29 : 1033-1039, 2019

66. Tao S, Zhang T, Wang K, Xie F, Ni L, Mei Z, et al : Identification of the risk factors in perioperative respiratory adverse events in children under general anesthesia and the development of a predictive model. Transl Pediatr 10 : 1877-1882, 2021

67. Thomas K, Hughes C, Johnson D, Das S : Anesthesia for surgery related to craniosynostosis: a review. Part 1. Paediatr Anaesth 22 : 1033-1041, 2012

68. Tremper KK, Mace JJ, Gombert JM, Tremper TT, Adams JF, Bagian JP : Design of a novel multifunction decision support display for anesthesia care: AlertWatch® OR. BMC Anesthesiol 18 : 16, 2018

69. van Lindert EJ, Arts S, Blok LM, Hendriks MP, Tielens L, van Bilsen M, et al : Intraoperative complications in pediatric neurosurgery: review of 1807 cases. J Neurosurg Pediatr 18 : 363-371, 2016

70. West N, Van Heusden K, Görges M, Brodie S, Rollinson A, Petersen CL, et al : Design and evaluation of a closed-loop anesthesia system with robust control and safety system. Anesth Analg 127 : 883-894, 2018

71. Wijnberge M, Geerts BF, Hol L, Lemmers N, Mulder MP, Berge P, et al : Effect of a machine learning-derived early warning system for intraoperative hypotension vs standard care on depth and duration of intraoperative hypotension during elective noncardiac surgery: the HYPE randomized clinical trial. JAMA 323 : 1052-1060, 2020

72. Wingert T, Lee C, Cannesson M : Machine learning, deep learning, and closed loop devices-anesthesia delivery. Anesthesiol Clin 39 : 565-581, 2021

73. Xue B, Li D, Lu C, King CR, Wildes T, Avidan MS, et al : Use of machine learning to develop and evaluate models using preoperative and intraoperative data to identify risks of postoperative complications. JAMA Netw Open 4 : e212240, 2021

74. Yang L, Li S, Zhong L, Qiu L, Xie L, Chen L : VACTERL association complicated with multiple airway abnormalities: a case report. Medicine (Baltimore) 98 : e17413, 2019

75. Yoon SB, Park JB, Kang P, Jang YE, Kim EH, Lee JH, et al : Electroencephalography and anesthetic depth in children under 2 years of age: a prospective observational study. Paediatr Anaesth 35 : 294-301, 2025

76. Yuan I, Chao JY, Kurth CD, Missett R, Cornelissen L : Intraoperative EEG monitoring in pediatric anesthesia. Curr Anesthesiol Rep 13 : 135-142, 2023

77. Yuan I, Xu T, Kurth CD : Using electroencephalography (EEG) to guide propofol and sevoflurane dosing in pediatric anesthesia. Anesthesiol Clin 38 : 709-725, 2020

78. Zhang Q, Shen F, Wei Q, Liu H, Li B, Zhang Q, et al : Development and validation of a risk nomogram model for perioperative respiratory adverse events in children undergoing airway surgery: an observational prospective cohort study. Risk Manag Healthc Policy 15 : 1-12, 2022

79. Zhou M, Xu WY, Xu S, Zang QL, Li Q, Tan L, et al : Prediction of endotracheal tube size in pediatric patients: development and validation of machine learning models. Front Pediatr 10 : 970646, 2022