A machine-learning algorithm for neonatal seizure recognition: a multicentre, randomised, controlled trial

Nursing Initiation of Rapid Electroencephalography Point-of-Care Monitoring: Lessons From the Pioneer Summit

ABSTRACT

BACKGROUND: Status epilepticus is an emergency, and applying electroencephalography (EEG) monitoring is an important part of diagnosing and treating seizure. The use of rapidly applied limited array continuous EEG (rapid EEG) has become technologically feasible in recent years. Nurse-led protocols using rapid EEG as a point-of-care monitor are increasingly being adopted. METHODS: A virtual summit meeting of nurses and physicians was convened to discuss various technological and practical aspects of rapid EEG, including the use of nurse-led protocols using rapid EEG. After oral presentations, participants responded to a survey indicating their level of agreement with key position statements. RESULTS: From the 52 participants who participated in the 2-hour summit, there was a strong agreement with the statement "Bedside nurses can start point-of-care EEG with automated seizure alert software to provide more informed care," with a median Likert score of 5 (completely agree) and an interquartile range of 4 to 5. CONCLUSION: Using rapid EEG to monitor for seizure is a valid and valuable method that falls within the nursing domain. Nurse-driven protocols may provide the opportunity to enhance patient care through early identification of seizures.

Transforming Healthcare in Low-Resource Settings With Artificial Intelligence: Recent Developments and Outcomes

BACKGROUND

Artificial intelligence now encompasses technologies like machine learning, natural language processing, and robotics, allowing machines to undertake complex tasks traditionally done by humans. AI's application in healthcare has led to advancements in diagnostic tools, predictive analytics, and surgical precision.

AIM

This comprehensive review aims to explore the transformative impact of AI across diverse healthcare domains, highlighting its applications, advancements, challenges, and contributions to enhancing patient care.

METHODOLOGY

A comprehensive literature search was conducted across multiple databases, covering publications from 2014 to 2024. Keywords related to AI applications in healthcare were used to gather data, focusing on studies exploring AI's role in medical specialties.

RESULTS

AI has demonstrated substantial benefits across various fields of medicine. In cardiology, it aids in automated image interpretation, risk prediction, and the management of cardiovascular diseases. In oncology, AI enhances cancer detection, treatment planning, and personalized drug selection. Radiology benefits from improved image analysis and diagnostic accuracy, while critical care sees advancements in patient triage and resource optimization. AI's integration into pediatrics, surgery, public health, neurology, pathology, and mental health has similarly shown significant improvements in diagnostic precision, personalized treatment, and overall patient care. The implementation of AI in low-resource settings has been particularly impactful, enhancing access to advanced diagnostic tools and treatments.

CONCLUSION

AI is rapidly changing the healthcare industry by greatly increasing the accuracy of diagnoses, streamlining treatment plans, and improving patient outcomes across a variety of medical specializations. This review underscores AI's transformative potential, from early disease detection to personalized treatment plans, and its ability to augment healthcare delivery, particularly in resource-limited settings.

Incidence and Predictors of Later Epilepsy in Neonates with Encephalopathy: The Impact of Electrographic Seizures

OBJECTIVES

To determine the incidence of later epilepsy in full-term infants with neonatal encephalopathy (NE) who undergo continuous electroencephalography (cEEG) monitoring in the neonatal period and to identify potential predictors of later epilepsy both in infants with and without electrographic neonatal seizures (ENS).

METHODS

This was a retrospective observational study performed at Cork University Maternity Hospital, Cork, Ireland, between 2003 and 2019. All term infants with NE had a minimum of 2 h of cEEG monitoring in the neonatal period. ENS were identified via cEEG monitoring. Pediatric medical charts were reviewed to determine if epilepsy developed after the neonatal period and to determine potential predictors of epilepsy in infants both with and without ENS.

RESULTS

Two hundred and eighty infants were included. The overall incidence rate of epilepsy was 17.55 per 1000 person-years (95% CI: 10.91 to 28.23). In infants with ENS (n = 82), the incidence rate was 39.27 per 1000 person-years (95% CI: 22.30 to 69.16). In infants without ENS (n = 198), the incidence rate was 7.54 per 1000 person-years (95% CI: 3.14 to 18.12). The incidence rate was significantly higher in the ENS group compared to the non-ENS group (p-value = 0.002). Several potential predictors for the development of later epilepsy were identified including infants delivered vaginally, low Apgar scores at 1 and 5 min, severe HIE diagnosis, presence of ENS, a severely abnormal EEG background and an abnormal brain MRI.

SIGNIFICANCE

Following NE, term infants are at risk of epilepsy with a significantly higher incidence rate in infants who experience ENS compared to those who did not. Close follow-up is required in both groups well into the childhood period.

PLAIN LANGUAGE SUMMARY

This study aimed to determine the occurrence of epilepsy in children who were monitored for seizures in the newborn period. The occurrence of epilepsy was higher in infants who experienced seizures in the newborn period compared to those who did not. Several potential predictors of later epilepsy were identified in both groups of infants (those with and without seizures in the newborn period). Both groups of infants require close follow-up in childhood.

Scaling convolutional neural networks achieves expert level seizure detection in neonatal EEG

Neonatal seizures require urgent treatment, but often go undetected without expert EEG monitoring. We have developed and validated a seizure detection model using retrospective EEG data from 332 neonates. A convolutional neural network was trained and tested on over 50,000 hours (n = 202) of annotated single-channel EEG containing 12,402 seizure events. This model was then validated on two independent multi-reviewer datasets (n = 51 and n = 79). Increasing data and model size improved performance: Matthews correlation coefficient (MCC) and Pearson's correlation (r) increased by up to 50% (15%) with data (model) scaling. The largest model (21m parameters) achieved state-of-the-art on an open-access dataset (MCC = 0.764, r = 0.824, and AUC = 0.982). This model also attained expert-level performance on both validation sets, a first in this field, with no significant difference in inter-rater agreement when the model replaces an expert (∣Δκ∣ < 0.094, p > 0.05).

A Comparison of Automatically Extracted Quantitative EEG Features for Seizure Risk Stratification in Neonatal Encephalopathy

PURPOSE

Seizures occur in up to 40% of neonates with neonatal encephalopathy. Earlier identification of seizures leads to more successful seizure treatment, but is often delayed because of limited availability of continuous EEG monitoring. Clinical variables poorly stratify seizure risk, and EEG use to stratify seizure risk has previously been limited by need for manual review and artifact exclusion. The goal of this study is to compare the utility of automatically extracted quantitative EEG (qEEG) features for seizure risk stratification.

METHODS

We conducted a retrospective analysis of neonates with moderate-to-severe neonatal encephalopathy who underwent therapeutic hypothermia at a single center. The first 24 hours of EEG underwent automated artifact removal and qEEG analysis, comparing qEEG features for seizure risk stratification.

RESULTS

The study included 150 neonates and compared the 36 (23%) with seizures with those without. Absolute spectral power best stratified seizure risk with area under the curve ranging from 63% to 71%, followed by range EEG lower and upper margin, median and SD of the range EEG lower margin. No features were significantly more predictive in the hour before seizure onset. Clinical examination was not associated with seizure risk.

CONCLUSIONS

Automatically extracted qEEG features were more predictive than clinical examination in stratifying neonatal seizure risk during therapeutic hypothermia. qEEG represents a potential practical bedside tool to individualize intensity and duration of EEG monitoring and decrease time to seizure recognition. Future work is needed to refine and combine qEEG features to improve risk stratification.

Benefits and harms associated with the use of AI-related algorithmic decision-making systems by healthcare professionals: a systematic review

Background

Despite notable advancements in artificial intelligence (AI) that enable complex systems to perform certain tasks more accurately than medical experts, the impact on patient-relevant outcomes remains uncertain. To address this gap, this systematic review assesses the benefits and harms associated with AI-related algorithmic decision-making (ADM) systems used by healthcare professionals, compared to standard care.

Methods

In accordance with the PRISMA guidelines, we included interventional and observational studies published as peer-reviewed full-text articles that met the following criteria: human patients; interventions involving algorithmic decision-making systems, developed with and/or utilizing machine learning (ML); and outcomes describing patient-relevant benefits and harms that directly affect health and quality of life, such as mortality and morbidity. Studies that did not undergo preregistration, lacked a standard-of-care control, or pertained to systems that assist in the execution of actions (e.g., in robotics) were excluded. We searched MEDLINE, EMBASE, IEEE Xplore, and Google Scholar for studies published in the past decade up to 31 March 2024. We assessed risk of bias using Cochrane's RoB 2 and ROBINS-I tools, and reporting transparency with CONSORT-AI and TRIPOD-AI. Two researchers independently managed the processes and resolved conflicts through discussion. This review has been registered with PROSPERO (CRD42023412156) and the study protocol has been published.

Findings

Out of 2,582 records identified after deduplication, 18 randomized controlled trials (RCTs) and one cohort study met the inclusion criteria, covering specialties such as psychiatry, oncology, and internal medicine. Collectively, the studies included a median of 243 patients (IQR 124-828), with a median of 50.5% female participants (range 12.5-79.0, IQR 43.6-53.6) across intervention and control groups. Four studies were classified as having low risk of bias, seven showed some concerns, and another seven were assessed as having high or serious risk of bias. Reporting transparency varied considerably: six studies showed high compliance, four moderate, and five low compliance with CONSORT-AI or TRIPOD-AI. Twelve studies (63%) reported patient-relevant benefits. Of those with low risk of bias, interventions reduced length of stay in hospital and intensive care unit (10.3 vs. 13.0 days, p = 0.042; 6.3 vs. 8.4 days, p = 0.030), in-hospital mortality (9.0% vs. 21.3%, p = 0.018), and depression symptoms in non-complex cases (45.1% vs. 52.3%, p = 0.03). However, harms were frequently underreported, with only eight studies (42%) documenting adverse events. No study reported an increase in adverse events as a result of the interventions.

Interpretation

The current evidence on AI-related ADM systems provides limited insights into patient-relevant outcomes. Our findings underscore the essential need for rigorous evaluations of clinical benefits, reinforced compliance with methodological standards, and balanced consideration of both benefits and harms to ensure meaningful integration into healthcare practice.

Funding

This study did not receive any funding.

AI models in clinical neonatology: a review of modeling approaches and a consensus proposal for standardized reporting of model performance

Artificial intelligence (AI) is a rapidly advancing area with growing clinical applications in healthcare. The neonatal intensive care unit (NICU) produces large amounts of multidimensional data allowing AI and machine learning (ML) new avenues to improve early diagnosis, enhance monitoring, and provide highly-targeted treatment approaches. In this article, we review recent clinical applications of AI to important neonatal problems, including sepsis, retinopathy of prematurity, bronchopulmonary dysplasia, and others. For each clinical area, we highlight a variety of ML models published in the literature and examine the future role they may play at the bedside. While the development of these models is rapidly expanding, a fundamental understanding of model selection, development, and performance evaluation is crucial for researchers and healthcare providers alike. As AI plays an increasing role in daily practice, understanding the implications of AI design and performance will enable more effective implementation. We provide a comprehensive explanation of the AI development process and recommendations for a standardized performance metric framework. Additionally, we address critical challenges, including model generalizability, ethical considerations, and the need for rigorous performance monitoring to avoid model drift. Finally, we outline future directions, emphasizing the importance of collaborative efforts and equitable access to AI innovations.

Neuromonitoring in the ICU: noninvasive and invasive modalities for critically ill children and neonates

PURPOSE OF REVIEW

Critically ill children are at risk of neurologic dysfunction and acquiring primary and secondary brain injury. Close monitoring of cerebral function is crucial to prevent, detect, and treat these complications.

RECENT FINDINGS

A variety of neuromonitoring modalities are currently used in pediatric and neonatal ICUs. These include noninvasive modalities, such as electroencephalography, transcranial Doppler, and near-infrared spectroscopy, as well as invasive methods including intracranial pressure monitoring, brain tissue oxygen measurement, and cerebral microdialysis. Each modality offers unique insights into neurologic function, cerebral circulation, or metabolism to support individualized neurologic care based on a patient's own physiology. Utilization of these modalities in ICUs results in reduced neurologic injury and mortality and improved neurodevelopmental outcomes.

SUMMARY

Monitoring of neurologic function can significantly improve care of critically ill children. Additional research is needed to establish normative values in pediatric patients and to standardize the use of these modalities.

Analysis of the impact of deep learning know-how and data in modelling neonatal EEG

The performance gains achieved by deep learning models nowadays are mainly attributed to the usage of ever larger datasets. In this study, we present and contrast the performance gains that can be achieved via accessing larger high-quality datasets versus the gains that can be achieved from harnessing the latest deep learning architectural and training advances. Modelling neonatal EEG is particularly affected by the lack of publicly available large datasets. It is shown that greater performance gains can be achieved from harnessing the latest deep learning advances than using a larger training dataset when adopting AUC as a metric, whereas using AUC90 or AUC-PR as metrics greater performance gains are achieved from using a larger dataset than harnessing the latest deep learning advances. In all scenarios the best performance is obtained by combining both deep learning advances and larger datasets. A novel developed architecture is presented that outperforms the current state-of-the-art model for the task of neonatal seizure detection. A novel method to fine-tune the presented model towards site-specific settings based on pseudo labelling is also outlined. The code and the weights of the model are made publicly available for benchmarking future model performances for neonatal seizure detection.

TATPat based explainable EEG model for neonatal seizure detection

The most cost-effective data collection method is electroencephalography (EEG) to obtain meaningful information about the brain. Therefore, EEG signal processing is very important for neuroscience and machine learning (ML). The primary objective of this research is to detect neonatal seizures and explain these seizures using the new version of Directed Lobish. This research uses a publicly available neonatal EEG signal dataset to get comparative results. In order to classify these EEG signals, an explainable feature engineering (EFE) model has been proposed. In this EFE model, there are four essential phases and these phases: (i) automaton and transformer-based feature extraction, (ii) feature selection deploying cumulative weight-based neighborhood component analysis (CWNCA), (iii) the Directed Lobish (DLob) and Causal Connectome Theory (CCT)-based explainable result generation and (iv) classification deploying t algorithm-based support vector machine (tSVM). In the first phase, we have used a channel transformer to get channel numbers and these values have been divided into three levels and these levels are named (1) high, (2) medium and (3) low. By utilizing these levels, we have created an automaton and this automaton has three nodes (each node defines each level). In the feature extraction phase, transition tables of these nodes has been extracted. Therefore, the proposed feature extraction function is termed Triple Nodes Automaton-based Transition table Pattern (TATPat). The used EEG signal dataset contains 19 channels and there are 9 (= 32) connection in the defined automaton. Thus, the presented TATPat extracts 3249 (= 19 × 19 × 9) features from each EEG segment. To choose the most informative features of these 3249 features, a new feature selector which is CWNCA has been applied. By cooperating findings of this feature selector and the presented DLob, the explainable results have been obtained. The last phase is the classification phase and to get high classification performance from this phase, an ensemble classifier (tSVM) has been presented and the classification results have been obtained using two validation techniques which are 10-fold cross-validation (CV) and leave-one subject-out (LOSO) CV. The proposed EFE model generates a DLob string and by using this string, the explainable results have been obtained. Moreover, the presented EFE model attained 99.15% and 76.37% classification accuracy deploying 10-fold and LOSO CVs respectively. According to the classification performances, the recommended TATPat-based EFE is a good model at EEG signal classification. Also, the presented TATPat-based EFE model is a good model for explainable artificial intelligence (XAI) since TTPat-based EFE is cooperating by the DLob.

Machine learning models of cerebral oxygenation (rcSO2) for brain injury detection in neonates with hypoxic-ischaemic encephalopathy

The present study was designed to test the potential utility of regional cerebral oxygen saturation (rcSO2) in detecting term infants with brain injury. The study also examined whether quantitative rcSO2 features are associated with grade of hypoxic ischaemic encephalopathy (HIE). We analysed 58 term infants with HIE (>36 weeks of gestational age) enrolled in a prospective observational study. All newborn infants had a period of continuous rcSO2 monitoring and magnetic resonance imaging (MRI) assessment during the first week of life. rcSO2 Signals were pre-processed and quantitative features were extracted. Machine-learning and deep-learning models were developed to detect adverse outcome (brain injury on MRI or death in the first week) using the leave-one-out cross-validation approach and to assess the association between rcSO2 and HIE grade (modified Sarnat - at 1 h). The machine-learning model (rcSO2 excluding prolonged relative desaturations) significantly detected infant MRI outcome or death in the first week of life [area under the curve (AUC) = 0.73, confidence interval (CI) = 0.59-0.86, Matthew's correlation coefficient = 0.35]. In agreement, deep learning models detected adverse outcome with an AUC = 0.64, CI = 0.50-0.79. We also report a significant association between rcSO2 features and HIE grade using a machine learning approach (AUC = 0.81, CI = 0.73-0.90). We conclude that automated analysis of rcSO2 using machine learning methods in term infants with HIE was able to determine, with modest accuracy, infants with adverse outcome. De novo approaches to signal analysis of NIRS holds promise to aid clinical decision making in the future. KEY POINTS: Hypoxic-induced neonatal brain injury contributes to both short- and long-term functional deficits. Non-invasive continuous monitoring of brain oxygenation using near-infrared- spectroscopy offers a potential new insight to the development of serious injury. In this study, characteristics of the NIRS signal were summarised using either predefined features or data-driven feature extraction, both were combined with a machine learning approach to predict short-term brain injury. Using data from a cohort of term infants with hypoxic ischaemic encephalopathy, the present study illustrates that automated analysis of regional cerebral oxygen saturation rcSO2, using either machine learning or deep learning methods, was able to determine infants with adverse outcome.

Real-Time Non-Invasive Imaging and Detection of Spreading Depolarizations through EEG: An Ultra-Light Explainable Deep Learning Approach

A core aim of neurocritical care is to prevent secondary brain injury. Spreading depolarizations (SDs) have been identified as an important independent cause of secondary brain injury. SDs are usually detected using invasive electrocorticography recorded at high sampling frequency. Recent pilot studies suggest a possible utility of scalp electrodes generated electroencephalogram (EEG) for non-invasive SD detection. However, noise and attenuation of EEG signals makes this detection task extremely challenging. Previous methods focus on detecting temporal power change of EEG over a fixed high-density map of scalp electrodes, which is not always clinically feasible. Having a specialized spectrogram as an input to the automatic SD detection model, this study is the first to transform SD identification problem from a detection task on a 1-D time-series wave to a task on a sequential 2-D rendered imaging. This study presented a novel ultra-light-weight multi-modal deep-learning network to fuse EEG spectrogram imaging and temporal power vectors to enhance SD identification accuracy over each single electrode, allowing flexible EEG map and paving the way for SD detection on ultra-low-density EEG with variable electrode positioning. Our proposed model has an ultra-fast processing speed (<0.3 sec). Compared to the conventional methods (2 hours), this is a huge advancement towards early SD detection and to facilitate instant brain injury prognosis. Seeing SDs with a new dimension - frequency on spectrograms, we demonstrated that such additional dimension could improve SD detection accuracy, providing preliminary evidence to support the hypothesis that SDs may show implicit features over the frequency profile.

Orphan and paediatric medical devices in Europe: recommendations to support their availability for on-label and off-label clinical indications

INTRODUCTION

The Medical Device Regulation (EU)745/2017, increased the regulatory requirements and thus the time and the cost associated with marketing medical devices. For a majority of medical device manufacturers, this has lead to reconsiderations of their product portfolio. The risk of important or essential devices being withdrawn is particularly relevant for pediatric patients and other rare disease patients where limited numbers of devices can be sold and hence the investment needed may not be recovered. This generates critical challenges and opportunities from a regulatory and public health perspective.

AREAS COVERED

This paper is based upon the experience of the authors who contributed to working groups, guidance development and research related to orphan and pediatric devices. We examine the use of medical devices in orphan and pediatric conditions, the relevant aspects of regulations and associated guidance, and we suggest possible policy and practice interventions to ensure the continued availability of essential devices for children and people with rare diseases.

EXPERT OPINION

We recommend a more proactive approach to identifying devices at risk and essential devices, increasing the use of exceptional market approvals, expanding the role of expert panels, engaging with the rare disease communities and supporting registries and standards.

Electrographic monitoring for seizure detection in the neonatal unit: current status and future direction

Neonatal neurocritical intensive care is dedicated to safeguarding the newborn brain by prioritising clinical practices that promote early identification, diagnosis and treatment of brain injuries. The most common newborn neurological emergency is neonatal seizures, which may also be the initial clinical indication of neurological disease. A high seizure burden in the newborn period independently contributes to increased mortality and morbidity. The majority of seizures in newborns are subclinical (without clinical presentation), and hence identification may be difficult. Neuromonitoring techniques most frequently used to monitor brain wave activity include conventional electroencephalography (cEEG) or amplitude-integrated EEG (aEEG). cEEG with video is the gold standard for diagnosing and treating seizures. Many neonatal units do not have access to cEEG, and frequently those that do, have little access to real-time interpretation of monitoring. IMPACT: EEG monitoring is of no benefit to an infant without expert interpretation. Whilst EEG is a reliable cot-side tool and of diagnostic and prognostic use, both conventional EEG and amplitude-integrated EEG have strengths and limitations, including sensitivity to seizure activity and ease of interpretation. Automated seizure detection requires a sensitive and specific algorithm that can interpret EEG in real-time and identify seizures, including their intensity and duration.

Free access via computational cloud to deep learning-based EEG assessment in neonatal hypoxic-ischemic encephalopathy: revolutionary opportunities to overcome health disparities

In this issue of Pediatric Research, Kota et al. evaluate a novel monitoring visual trend using deep-learning - Brain State of the Newborn (BSN)- based EEG as a bedside marker for severity of the encephalopathy in 46 neonates with hypoxic-ischemic encephalopathy (HIE) compared with healthy infants. Early BSN distinguished between normal and abnormal outcome, and correlated with the Total Sarnat Score.

Quantitative EEG and prediction of outcome in neonatal encephalopathy: a review

Electroencephalogram (EEG) is an important biomarker for neonatal encephalopathy (NE) and has significant predictive value for brain injury and neurodevelopmental outcomes. Quantitative analysis of EEG involves the representation of complex EEG data in an objective, reproducible and scalable manner. Quantitative EEG (qEEG) can be derived from both a limited channel EEG (as available during amplitude integrated EEG) and multi-channel conventional EEG. It has the potential to enable bedside clinicians to monitor and evaluate details of cortical function without the necessity of continuous expert input. This is particularly useful in NE, a dynamic and evolving condition. In these infants, continuous, detailed evaluation of cortical function at the bedside is a valuable aide to management especially in the current era of therapeutic hypothermia and possible upcoming neuroprotective therapies. This review discusses the role of qEEG in newborns with NE and its use in informing monitoring and therapy, along with its ability to predict imaging changes and short and long-term neurodevelopmental outcomes. IMPACT: Quantitative representation of EEG data brings the evaluation of continuous brain function, from the neurophysiology lab to the NICU bedside and has a potential role as a biomarker for neonatal encephalopathy. Clinical and research applications of quantitative EEG in the newborn are rapidly evolving and a wider understanding of its utility is valuable. This overview summarizes the role of quantitative EEG at different timepoints, its relevance to management and its predictive value for short- and long-term outcomes in neonatal encephalopathy.

Clinical outcome prediction with an automated EEG trend, Brain State of the Newborn, after perinatal asphyxia

OBJECTIVE

To evaluate the utility of a fully automated deep learning -based quantitative measure of EEG background, Brain State of the Newborn (BSN), for early prediction of clinical outcome at four years of age.

METHODS

The EEG monitoring data from eighty consecutive newborns was analyzed using the automatically computed BSN trend. BSN levels during the first days of life (a of total 5427 hours) were compared to four clinical outcome categories: favorable, cerebral palsy (CP), CP with epilepsy, and death. The time dependent changes in BSN-based prediction for different outcomes were assessed by positive/negative predictive value (PPV/NPV) and by estimating the area under the receiver operating characteristic curve (AUC).

RESULTS

The BSN values were closely aligned with four visually determined EEG categories (p < 0·001), as well as with respect to clinical milestones of EEG recovery in perinatal Hypoxic Ischemic Encephalopathy (HIE; p < 0·003). Favorable outcome was related to a rapid recovery of the BSN trend, while worse outcomes related to a slow BSN recovery. Outcome predictions with BSN were accurate from 6 to 48 hours of age: For the favorable outcome, the AUC ranged from 95 to 99% (peak at 12 hours), and for the poor outcome the AUC ranged from 96 to 99% (peak at 12 hours). The optimal BSN levels for each PPV/NPV estimate changed substantially during the first 48 hours, ranging from 20 to 80.

CONCLUSIONS

We show that the BSN provides an automated, objective, and continuous measure of brain activity in newborns.

SIGNIFICANCE

The BSN trend discloses the dynamic nature that exists in both cerebral recovery and outcome prediction, supports individualized patient care, rapid stratification and early prognosis.

Randomised controlled trials evaluating artificial intelligence in clinical practice: a scoping review

This scoping review of randomised controlled trials on artificial intelligence (AI) in clinical practice reveals an expanding interest in AI across clinical specialties and locations. The USA and China are leading in the number of trials, with a focus on deep learning systems for medical imaging, particularly in gastroenterology and radiology. A majority of trials (70 [81%] of 86) report positive primary endpoints, primarily related to diagnostic yield or performance; however, the predominance of single-centre trials, little demographic reporting, and varying reports of operational efficiency raise concerns about the generalisability and practicality of these results. Despite the promising outcomes, considering the likelihood of publication bias and the need for more comprehensive research including multicentre trials, diverse outcome measures, and improved reporting standards is crucial. Future AI trials should prioritise patient-relevant outcomes to fully understand AI's true effects and limitations in health care.

Temporal evolution of electrographic seizures in newborn infants with hypoxic-ischaemic encephalopathy requiring therapeutic hypothermia: a secondary analysis of the ANSeR studies

BACKGROUND

Despite extensive research on neonatal hypoxic-ischaemic encephalopathy, detailed information about electrographic seizures during active cooling and rewarming of therapeutic hypothermia is sparse. We aimed to describe temporal evolution of seizures and determine whether there is a correlation of seizure evolution with 2-year outcome.

METHODS

This secondary analysis included newborn infants recruited from eight European tertiary neonatal intensive care units for two multicentre studies (a randomised controlled trial [NCT02431780] and an observational study [NCT02160171]). Infants were born at 36+0 weeks of gestation with moderate or severe hypoxic-ischaemic encephalopathy and underwent therapeutic hypothermia with prolonged conventional video-electroencephalography (EEG) monitoring for 10 h or longer from the start of rewarming. Seizure burden characteristics were calculated based on electrographic seizures annotations: hourly seizure burden (minutes of seizures within an hour) and total seizure burden (minutes of seizures within the entire recording). We categorised infants into those with electrographic seizures during active cooling only, those with electrographic seizures during cooling and rewarming, and those without seizures. Neurodevelopmental outcomes were determined using the Bayley's Scales of Infant and Toddler Development, Third Edition (BSID-III), the Griffiths Mental Development Scales (GMDS), or neurological assessment. An abnormal outcome was defined as death or neurodisability at 2 years. Neurodisability was defined as a composite score of 85 or less on any subscales for BSID-III, a total score of 87 or less for GMDS, or a diagnosis of cerebral palsy (dyskinetic cerebral palsy, spastic quadriplegia, or mixed motor impairment) or epilepsy.

FINDINGS

Of 263 infants recruited between Jan 1, 2011, and Feb 7, 2017, we included 129 infants: 65 had electrographic seizures (43 during active cooling only and 22 during and after active cooling) and 64 had no seizures. Compared with infants with seizures during active cooling only, those with seizures during and after active cooling had a longer seizure period (median 12 h [IQR 3-28] vs 68 h [35-86], p<0·0001), more seizures (median 12 [IQR 5-36] vs 94 [24-134], p<0·0001), and higher total seizure burden (median 69 min [IQR 22-104] vs 167 min [54-275], p=0·0033). Hourly seizure burden peaked at about 20-24 h in both groups, and infants with seizures during and after active cooling had a secondary peak at 85 h of age. When combined, worse EEG background (major abnormalities and inactive background) at 12 h and 24 h were associated with the seizure group: compared with infants with a better EEG background (normal, mild, or moderate abnormalities), infants with a worse EEG background were more likely to have seizures after cooling at 12 h (13 [54%] of 24 vs four [14%] of 28; odds ratio 7·09 [95% CI 1·88-26·77], p=0·0039) and 24 h (14 [56%] of 25 vs seven [18%] of 38; 5·64 [1·81-17·60], p=0·0029). There was a significant relationship between EEG grade at 12 h (four categories) and seizure group (p=0·020). High total seizure burden was associated with increased odds of an abnormal outcome at 2 years of age (odds ratio 1·007 [95% CI 1·000-1·014], p=0·046), with a medium negative correlation between total seizure burden and BSID-III cognitive score (rS=-0·477, p=0·014, n=26).

INTERPRETATION

Overall, half of infants with hypoxic-ischaemic encephalopathy had electrographic seizures and a third of those infants had seizures beyond active cooling, with worse outcomes. These results raise the importance of prolonged EEG monitoring of newborn infants with hypoxic-ischaemic encephalopathy not only during active cooling but throughout the rewarming phase and even longer when seizures are detected.

FUNDING

Wellcome Trust, Science Foundation Ireland, and the Irish Health Research Board.

Towards trustworthy seizure onset detection using workflow notes

A major barrier to deploying healthcare AI is trustworthiness. One form of trustworthiness is a model's robustness across subgroups: while models may exhibit expert-level performance on aggregate metrics, they often rely on non-causal features, leading to errors in hidden subgroups. To take a step closer towards trustworthy seizure onset detection from EEG, we propose to leverage annotations that are produced by healthcare personnel in routine clinical workflows-which we refer to as workflow notes-that include multiple event descriptions beyond seizures. Using workflow notes, we first show that by scaling training data to 68,920 EEG hours, seizure onset detection performance significantly improves by 12.3 AUROC (Area Under the Receiver Operating Characteristic) points compared to relying on smaller training sets with gold-standard labels. Second, we reveal that our binary seizure onset detection model underperforms on clinically relevant subgroups (e.g., up to a margin of 6.5 AUROC points between pediatrics and adults), while having significantly higher FPRs (False Positive Rates) on EEG clips showing non-epileptiform abnormalities (+19 FPR points). To improve model robustness to hidden subgroups, we train a multilabel model that classifies 26 attributes other than seizures (e.g., spikes and movement artifacts) and significantly improve overall performance (+5.9 AUROC points) while greatly improving performance among subgroups (up to +8.3 AUROC points) and decreasing false positives on non-epileptiform abnormalities (by 8 FPR points). Finally, we find that our multilabel model improves clinical utility (false positives per 24 EEG hours) by a factor of 2×.

Acute symptomatic seizures in newborns: a narrative review

Acute symptomatic seizures are the main sign of neurological dysfunction in newborns. This is linked to the unique characteristics of the neonatal brain, making it hyperexcitable compared to older ages, and to the common occurrence of some forms of acquired brain injury, namely hypoxic-ischemic encephalopathy. In this narrative review we will provide an overview of neonatal seizures definition, their main underlying etiologies, diagnostic work-up and differential diagnoses, and will discuss about therapeutic options and prognostic outlook. The latest publications from the ILAE Task Force on Neonatal Seizures will be presented and discussed. Of note, they highlight the current lack of robust evidence in this field of clinical neurology. We will also report on specificities pertaining to low-and-middle income countries in terms of incidence, main etiologies and diagnosis. The possibilities offered by telemedicine and automated seizures detection will also be summarized in order to provide a framework for future directions in seizures diagnosis and management with a global perspective. Many challenges and opportunities for improving identification, monitoring and treatment of acute symptomatic seizures in newborns exist. All current caveats potentially represent different lines of research with the aim to provide better care and reach a deeper understanding of this important topic of neonatal neurology.

Relationship of Neonatal Seizure Burden Before Treatment and Response to Initial Antiseizure Medication

OBJECTIVE

To assess among a cohort of neonates with hypoxic-ischemic encephalopathy (HIE) the association of pretreatment maximal hourly seizure burden and total seizure duration with successful response to initial antiseizure medication (ASM).

STUDY DESIGN

This was a retrospective review of data collected from infants enrolled in the HEAL Trial (NCT02811263) between January 25, 2017, and October 9, 2019. We evaluated a cohort of neonates born at ≥36 weeks of gestation with moderate-to-severe HIE who underwent continuous electroencephalogram monitoring and had acute symptomatic seizures. Poisson regression analyzed associations between (1) pretreatment maximal hourly seizure burden, (2) pretreatment total seizure duration, (3) time from first seizure to initial ASM, and (4) successful response to initial ASM.

RESULTS

Among 39 neonates meeting inclusion criteria, greater pretreatment maximal hourly seizure burden was associated with lower chance of successful response to initial ASM (adjusted relative risk for each 5-minute increase in seizure burden 0.83, 95% CI 0.69-0.99). There was no association between pretreatment total seizure duration and chance of successful response. Shorter time-to-treatment was paradoxically associated with lower chance of successful response to treatment, although this difference was small in magnitude (relative risk 1.007, 95% CI 1.003-1.010).

CONCLUSIONS

Maximal seizure burden may be more important than other, more commonly used measures in predicting response to acute seizure treatments.

Transforming neonatal care with artificial intelligence: challenges, ethical consideration, and opportunities

Artificial intelligence (AI) offers tremendous potential to transform neonatology through improved diagnostics, personalized treatments, and earlier prevention of complications. However, there are many challenges to address before AI is ready for clinical practice. This review defines key AI concepts and discusses ethical considerations and implicit biases associated with AI. Next we will review literature examples of AI already being explored in neonatology research and we will suggest future potentials for AI work. Examples discussed in this article include predicting outcomes such as sepsis, optimizing oxygen therapy, and image analysis to detect brain injury and retinopathy of prematurity. Realizing AI's potential necessitates collaboration between diverse stakeholders across the entire process of incorporating AI tools in the NICU to address testability, usability, bias, and transparency. With multi-center and multi-disciplinary collaboration, AI holds tremendous potential to transform the future of neonatology.

The critical role of technologies in neonatal care

Neonatal care has made significant advances in the last few decades. As a result, mortality and morbidity in high-risk infants, such as extremely preterm infants or those infants with birth-related brain injury, has reduced significantly. Many of these advances have been facilitated or delivered through development of medical technologies allowing clinical teams to be better supported with the care they deliver or provide new therapies and diagnostics to improve management. The delivery of neonatal intensive care requires the provision of medical technologies that are easy to use, reliable, accurate and ideally developed for the unique needs of the newborn population. Many technologies have been developed and commercialised following adult trials without ever being studied in neonatal patients despite the unique characteristics of this population. Increasingly, funders and industry are recognising this major challenge which has resulted in initiatives to develop new ideas from concept through to clinical care. This review explores some of the key medical technologies used in neonatal care and the evidence to support their adoption to improve outcomes. A number of devices have yet to realise their full potential and will require further development to optimise and find their ideal target population and clinical benefit. Examples of emerging technologies, which may soon become more widely used, are also discussed. As neonatal care relies more on medical technologies, we need to be aware of the impact on care pathways, especially from a human factors approach, the associated costs and subsequent benefits to patients alongside the supporting evidence.

The past, current, and future of neonatal intensive care units with artificial intelligence: a systematic review

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.

Electroencephalography Quality and Application Times in a Pediatric Emergency Department Setting: A Feasibility Study

BACKGROUND

Status epilepticus is the most common neurological emergency presenting to pediatric emergency departments. Nonconvulsive status epilepticus can be extremely challenging to diagnose, however, requiring electroencephalographic (EEG) confirmation for definitive diagnosis. We aimed to determine the feasibility of achieving a good-quality pediatric EEG recording within 20 minutes of presentation to the emergency department.

METHODS

Single-center prospective feasibility study in Cork University Hospital, Ireland, between July 2021 and June 2022. Two-channel continuous EEG was recorded from children (1) aged <16 years and (2) with Glasgow Coma Scale <11 or a reduction in baseline Glasgow Coma Scale in the case of a child with a neurodisability.

RESULTS

Twenty patients were included. The median age at presentation was 65.8 months (interquartile range, 23.2 to 119.0); 50% had a background diagnosis of epilepsy. The most common reason for EEG monitoring was status epilepticus (85%) followed by suspected nonconvulsive status (10%) and reduced consciousness of unknown etiology (5%). The mean length of recording was 93.1 minutes (S.D. 47.4). The mean time to application was 41.3 minutes (S.D. 11.7). The mean percent of artifact in all recordings was 19.3% (S.D. 15.9). Thirteen (65%) EEGs had <25% artifact. Artifact was higher in cases in which active airway management was ongoing.

CONCLUSIONS

EEG monitoring can be achieved in a pediatric emergency department setting within one hour of presentation. Overall, artifact percentage was low outside of periods of airway manipulation. Future studies are required to determine its use in early seizure detection and its support role in clinical decision-making in these patients.

Remote Monitoring for Seizures During Therapeutic Hypothermia in Neonates With Hypoxic-Ischemic Encephalopathy

Importance

Neonates with hypoxic-ischemic encephalopathy (HIE) undergoing therapeutic hypothermia (TH) frequently experience seizures, which are associated with adverse outcomes. Efforts to rapidly identify seizures and reduce seizure burden may positively change neurologic and neurodevelopmental outcomes.

Objective

To describe the onset, treatment, and evolution of seizures in a large cohort of newborns with HIE during TH assisted by a telehealth model and remote neuromonitoring approach.

Design, Setting, and Participants

This was a prospective, observational, multicenter cohort study performed between July 2017 and December 2021 in 32 hospitals in Brazil. Participants were newborns with HIE meeting eligibility criteria and receiving TH. Data were analyzed from November 2022 to April 2023.

Exposure

Infants with HIE receiving TH were remotely monitored with 3-channel amplitude-integrated electroencephalography (aEEG) including raw tracing and video imaging, and bedside clinicians received assistance from trained neonatologists and neurologists.

Main Outcomes and Measures

Data on modified Sarnat examination, presence, timing and seizure type, aEEG background activity, sleep-wake cycling, and antiepileptic drugs used were collected. Descriptive statistical analysis was used with independent t test, χ2, Mann-Whitney test, and post hoc analyses applied for associations.

Results

A total of 872 cooled newborns were enrolled; the median (IQR) gestational age was 39 (38-40) weeks, 518 (59.4%) were male, and 59 (6.8%) were classified as having mild encephalopathy by modified Sarnat examination, 504 (57.8%) as moderate, and 180 (20.6%) as severe. Electrographic seizures were identified in 296 newborns (33.9%), being only electrographic in 213 (71.9%) and clinical followed by electroclinical uncoupling in 50 (16.9%). Early abnormal background activity had a significant association with seizures. Infants with flat trace had the highest rate of seizures (58 infants [68.2%]) and the greatest association with the incidence of seizures (odds ratio [OR], 12.90; 95% CI, 7.57-22.22) compared with continuous normal voltage. The absence of sleep-wake cycling was also associated with a higher occurrence of seizures (OR, 2.22; 95% CI, 1.67-2.96). Seizure onset was most frequent between 6 and 24 hours of life (181 infants [61.1%]); however, seizure occurred in 34 infants (11.5%) during rewarming. A single antiepileptic drug controlled seizures in 192 infants (64.9%). The first line antiepileptic drug was phenobarbital in 294 (99.3%).

Conclusions and Relevance

In this cohort study of newborns with HIE treated with TH, electrographic seizure activity occurred in 296 infants (33.9%) and was predominantly electrographic. Seizure control was obtained with a single antiepileptic drug in 192 infants (64.9%). These findings suggest neonatal neurocritical care can be delivered at remote limited resource hospitals due to innovations in technology and telehealth.

Anti-seizure medications for neonates with seizures

BACKGROUND

Newborn infants are more prone to seizures than older children and adults. The neuronal injury caused by seizures in neonates often results in long-term neurodevelopmental sequelae. There are several options for anti-seizure medications (ASMs) in neonates. However, the ideal choice of first-, second- and third-line ASM is still unclear. Further, many other aspects of seizure management such as whether ASMs should be initiated for only-electrographic seizures and how long to continue the ASM once seizure control is achieved are elusive.

OBJECTIVES

1. To assess whether any ASM is more or less effective than an alternative ASM (both ASMs used as first-, second- or third-line treatment) in achieving seizure control and improving neurodevelopmental outcomes in neonates with seizures. We analysed EEG-confirmed seizures and clinically-diagnosed seizures separately. 2. To assess maintenance therapy with ASM versus no maintenance therapy after achieving seizure control. We analysed EEG-confirmed seizures and clinically-diagnosed seizures separately. 3. To assess treatment of both clinical and electrographic seizures versus treatment of clinical seizures alone in neonates.

SEARCH METHODS

We searched MEDLINE, Embase, CENTRAL, Epistemonikos and three databases in May 2022 and June 2023. These searches were not limited other than by study design to trials.

SELECTION CRITERIA

We included randomised controlled trials (RCTs) that included neonates with EEG-confirmed or clinically diagnosed seizures and compared (1) any ASM versus an alternative ASM, (2) maintenance therapy with ASM versus no maintenance therapy, and (3) treatment of clinical or EEG seizures versus treatment of clinical seizures alone.

DATA COLLECTION AND ANALYSIS

Two review authors assessed trial eligibility, risk of bias and independently extracted data. We analysed treatment effects in individual trials and reported risk ratio (RR) for dichotomous data, and mean difference (MD) for continuous data, with respective 95% confidence interval (CI). We used GRADE to assess the certainty of evidence.

MAIN RESULTS

We included 18 trials (1342 infants) in this review. Phenobarbital versus levetiracetam as first-line ASM in EEG-confirmed neonatal seizures (one trial) Phenobarbital is probably more effective than levetiracetam in achieving seizure control after first loading dose (RR 2.32, 95% CI 1.63 to 3.30; 106 participants; moderate-certainty evidence), and after maximal loading dose (RR 2.83, 95% CI 1.78 to 4.50; 106 participants; moderate-certainty evidence). However, we are uncertain about the effect of phenobarbital when compared to levetiracetam on mortality before discharge (RR 0.30, 95% CI 0.04 to 2.52; 106 participants; very low-certainty evidence), requirement of mechanical ventilation (RR 1.21, 95% CI 0.76 to 1.91; 106 participants; very low-certainty evidence), sedation/drowsiness (RR 1.74, 95% CI 0.68 to 4.44; 106 participants; very low-certainty evidence) and epilepsy post-discharge (RR 0.92, 95% CI 0.48 to 1.76; 106 participants; very low-certainty evidence). The trial did not report on mortality or neurodevelopmental disability at 18 to 24 months. Phenobarbital versus phenytoin as first-line ASM in EEG-confirmed neonatal seizures (one trial) We are uncertain about the effect of phenobarbital versus phenytoin on achieving seizure control after maximal loading dose of ASM (RR 0.97, 95% CI 0.54 to 1.72; 59 participants; very low-certainty evidence). The trial did not report on mortality or neurodevelopmental disability at 18 to 24 months. Maintenance therapy with ASM versus no maintenance therapy in clinically diagnosed neonatal seizures (two trials) We are uncertain about the effect of short-term maintenance therapy with ASM versus no maintenance therapy during the hospital stay (but discontinued before discharge) on the risk of repeat seizures before hospital discharge (RR 0.76, 95% CI 0.56 to 1.01; 373 participants; very low-certainty evidence). Maintenance therapy with ASM compared to no maintenance therapy may have little or no effect on mortality before discharge (RR 0.69, 95% CI 0.39 to 1.22; 373 participants; low-certainty evidence), mortality at 18 to 24 months (RR 0.94, 95% CI 0.34 to 2.61; 111 participants; low-certainty evidence), neurodevelopmental disability at 18 to 24 months (RR 0.89, 95% CI 0.13 to 6.12; 108 participants; low-certainty evidence) and epilepsy post-discharge (RR 3.18, 95% CI 0.69 to 14.72; 126 participants; low-certainty evidence). Treatment of both clinical and electrographic seizures versus treatment of clinical seizures alone in neonates (two trials) Treatment of both clinical and electrographic seizures when compared to treating clinical seizures alone may have little or no effect on seizure burden during hospitalisation (MD -1871.16, 95% CI -4525.05 to 782.73; 68 participants; low-certainty evidence), mortality before discharge (RR 0.59, 95% CI 0.28 to 1.27; 68 participants; low-certainty evidence) and epilepsy post-discharge (RR 0.75, 95% CI 0.12 to 4.73; 35 participants; low-certainty evidence). The trials did not report on mortality or neurodevelopmental disability at 18 to 24 months. We report data from the most important comparisons here; readers are directed to Results and Summary of Findings tables for all comparisons.

AUTHORS' CONCLUSIONS

Phenobarbital as a first-line ASM is probably more effective than levetiracetam in achieving seizure control after the first loading dose and after the maximal loading dose of ASM (moderate-certainty evidence). Phenobarbital + bumetanide may have little or no difference in achieving seizure control when compared to phenobarbital alone (low-certainty evidence). Limited data and very low-certainty evidence preclude us from drawing any reasonable conclusion on the effect of using one ASM versus another on other short- and long-term outcomes. In neonates who achieve seizure control after the first loading dose of phenobarbital, maintenance therapy compared to no maintenance ASM may have little or no effect on all-cause mortality before discharge, mortality by 18 to 24 months, neurodevelopmental disability by 18 to 24 months and epilepsy post-discharge (low-certainty evidence). In neonates with hypoxic-ischaemic encephalopathy, treatment of both clinical and electrographic seizures when compared to treating clinical seizures alone may have little or no effect on seizure burden during hospitalisation, all-cause mortality before discharge and epilepsy post-discharge (low-certainty evidence). All findings of this review apply only to term and late preterm neonates. We need well-designed RCTs for each of the three objectives of this review to improve the precision of the results. These RCTs should use EEG to diagnose seizures and should be adequately powered to assess long-term neurodevelopmental outcomes. We need separate RCTs evaluating the choice of ASM in preterm infants.

Quantitative analysis of high-frequency activity in neonatal EEG

OBJECTIVE

To determine the presence and potential utility of independent high-frequency activity recorded from scalp electrodes in the electroencephalogram (EEG) of newborns.

METHODS

We compare interburst intervals and continuous activity at different frequencies for EEGs retrospectively recorded at 256 Hz from 4 newborn groups: 1) 36 preterms (<32 weeks' gestational age, GA); 2) 12 preterms (32-37 weeks' GA); 3) 91 healthy full terms; 4) 15 full terms with hypoxic-ischemic encephalopathy (HIE). At 4 standard frequency bands (delta, 0.5-3 Hz; theta, 3-8 Hz; alpha, 8-15 Hz; beta, 15-30 Hz) and 3 higher-frequency bands (gamma1, 30-48 Hz; gamma2, 52-99 Hz; gamma3, 107-127 Hz), we compared power spectral densities (PSDs), quantitative features, and machine learning model performance. Feature selection and further machine learning methods were performed on one cohort.

RESULTS

We found significant (P < 0.01) differences in PSDs, quantitative analysis, and machine learning modelling at the higher-frequency bands. Machine learning models using only high-frequency features performed best in preterm groups 1 and 2 with a median (95% confidence interval, CI) Matthews correlation coefficient (MCC) of 0.71 (0.12-0.88) and 0.66 (0.36-0.76) respectively. Interburst interval-detector models using both high- and standard-bandwidths produced the highest median MCCs in all four groups. High-frequency features were largely independent of standard-bandwidth features, with only 11/84 (13.1%) of correlations statistically significant. Feature selection methods produced 7 to 9 high-frequency features in the top 20 feature set.

CONCLUSIONS

This is the first study to identify independent high-frequency activity in newborn EEG using in-depth quantitative analysis. Expanding the EEG bandwidths of analysis has the potential to improve both quantitative and machine-learning analysis, particularly in preterm EEG.

A Class-Imbalance Aware and Explainable Spatio-Temporal Graph Attention Network for Neonatal Seizure Detection

Seizures are the most prevalent clinical indication of neurological disorders in neonates. In this study, a class-imbalance aware and explainable deep learning approach based on Convolutional Neural Networks (CNNs) and Graph Attention Networks (GATs) is proposed for the accurate automated detection of neonatal seizures. The proposed model integrates the temporal information of EEG signals with the spatial information on the EEG channels through the graph representation of the multi-channel EEG segments. One-dimensional CNNs are used to automatically develop a feature set that accurately represents the differences between seizure and nonseizure epochs in the time domain. By employing GAT, the attention mechanism is utilized to emphasize the critical channel pairs and information flow among brain regions. GAT coefficients were then used to empirically visualize the important regions during the seizure and nonseizure epochs, which can provide valuable insight into the location of seizures in the neonatal brain. Additionally, to tackle the severe class imbalance in the neonatal seizure dataset using under-sampling and focal loss techniques are used. Overall, the final Spatio-Temporal Graph Attention Network (ST-GAT) outperformed previous benchmarked methods with a mean AUC of 96.6% and Kappa of 0.88, demonstrating its high accuracy and potential for clinical applications.

Automated Interpretation of Clinical Electroencephalograms Using Artificial Intelligence

Importance

Electroencephalograms (EEGs) are a fundamental evaluation in neurology but require special expertise unavailable in many regions of the world. Artificial intelligence (AI) has a potential for addressing these unmet needs. Previous AI models address only limited aspects of EEG interpretation such as distinguishing abnormal from normal or identifying epileptiform activity. A comprehensive, fully automated interpretation of routine EEG based on AI suitable for clinical practice is needed.

Objective

To develop and validate an AI model (Standardized Computer-based Organized Reporting of EEG-Artificial Intelligence [SCORE-AI]) with the ability to distinguish abnormal from normal EEG recordings and to classify abnormal EEG recordings into categories relevant for clinical decision-making: epileptiform-focal, epileptiform-generalized, nonepileptiform-focal, and nonepileptiform-diffuse.

Design, Setting, and Participants

In this multicenter diagnostic accuracy study, a convolutional neural network model, SCORE-AI, was developed and validated using EEGs recorded between 2014 and 2020. Data were analyzed from January 17, 2022, until November 14, 2022. A total of 30 493 recordings of patients referred for EEG were included into the development data set annotated by 17 experts. Patients aged more than 3 months and not critically ill were eligible. The SCORE-AI was validated using 3 independent test data sets: a multicenter data set of 100 representative EEGs evaluated by 11 experts, a single-center data set of 9785 EEGs evaluated by 14 experts, and for benchmarking with previously published AI models, a data set of 60 EEGs with external reference standard. No patients who met eligibility criteria were excluded.

Main Outcomes and Measures

Diagnostic accuracy, sensitivity, and specificity compared with the experts and the external reference standard of patients' habitual clinical episodes obtained during video-EEG recording.

Results

The characteristics of the EEG data sets include development data set (N = 30 493; 14 980 men; median age, 25.3 years [95% CI, 1.3-76.2 years]), multicenter test data set (N = 100; 61 men, median age, 25.8 years [95% CI, 4.1-85.5 years]), single-center test data set (N = 9785; 5168 men; median age, 35.4 years [95% CI, 0.6-87.4 years]), and test data set with external reference standard (N = 60; 27 men; median age, 36 years [95% CI, 3-75 years]). The SCORE-AI achieved high accuracy, with an area under the receiver operating characteristic curve between 0.89 and 0.96 for the different categories of EEG abnormalities, and performance similar to human experts. Benchmarking against 3 previously published AI models was limited to comparing detection of epileptiform abnormalities. The accuracy of SCORE-AI (88.3%; 95% CI, 79.2%-94.9%) was significantly higher than the 3 previously published models (P < .001) and similar to human experts.

Conclusions and Relevance

In this study, SCORE-AI achieved human expert level performance in fully automated interpretation of routine EEGs. Application of SCORE-AI may improve diagnosis and patient care in underserved areas and improve efficiency and consistency in specialized epilepsy centers.

Technical feasibility of multimodal imaging in neonatal hypoxic-ischemic encephalopathy from an ovine model to a human case series

Hypoxic-ischemic encephalopathy (HIE) secondary to perinatal asphyxia occurs when the brain does not receive enough oxygen and blood. A surrogate marker for "intact survival" is necessary for the successful management of HIE. The severity of HIE can be classified based on clinical presentation, including the presence of seizures, using a clinical classification scale called Sarnat staging; however, Sarnat staging is subjective, and the score changes over time. Furthermore, seizures are difficult to detect clinically and are associated with a poor prognosis. Therefore, a tool for continuous monitoring on the cot side is necessary, for example, an electroencephalogram (EEG) that noninvasively measures the electrical activity of the brain from the scalp. Then, multimodal brain imaging, when combined with functional near-infrared spectroscopy (fNIRS), can capture the neurovascular coupling (NVC) status. In this study, we first tested the feasibility of a low-cost EEG-fNIRS imaging system to differentiate between normal, hypoxic, and ictal states in a perinatal ovine hypoxia model. Here, the objective was to evaluate a portable cot-side device and perform autoregressive with extra input (ARX) modeling to capture the perinatal ovine brain states during a simulated HIE injury. So, ARX parameters were tested with a linear classifier using a single differential channel EEG, with varying states of tissue oxygenation detected using fNIRS, to label simulated HIE states in the ovine model. Then, we showed the technical feasibility of the low-cost EEG-fNIRS device and ARX modeling with support vector machine classification for a human HIE case series with and without sepsis. The classifier trained with the ovine hypoxia data labeled ten severe HIE human cases (with and without sepsis) as the "hypoxia" group and the four moderate HIE human cases as the "control" group. Furthermore, we showed the feasibility of experimental modal analysis (EMA) based on the ARX model to investigate the NVC dynamics using EEG-fNIRS joint-imaging data that differentiated six severe HIE human cases without sepsis from four severe HIE human cases with sepsis. In conclusion, our study showed the technical feasibility of EEG-fNIRS imaging, ARX modeling of NVC for HIE classification, and EMA that may provide a biomarker of sepsis effects on the NVC in HIE.

Neonatal EEG graded for severity of background abnormalities in hypoxic-ischaemic encephalopathy

This report describes a set of neonatal electroencephalogram (EEG) recordings graded according to the severity of abnormalities in the background pattern. The dataset consists of 169 hours of multichannel EEG from 53 neonates recorded in a neonatal intensive care unit. All neonates received a diagnosis of hypoxic-ischaemic encephalopathy (HIE), the most common cause of brain injury in full term infants. For each neonate, multiple 1-hour epochs of good quality EEG were selected and then graded for background abnormalities. The grading system assesses EEG attributes such as amplitude, continuity, sleep-wake cycling, symmetry and synchrony, and abnormal waveforms. Background severity was then categorised into 4 grades: normal or mildly abnormal EEG, moderately abnormal EEG, majorly abnormal EEG, and inactive EEG. The data can be used as a reference set of multi-channel EEG for neonates with HIE, for EEG training purposes, or for developing and evaluating automated grading algorithms.

Applications of Artificial Intelligence in Neonatology

The development of artificial intelligence methods has impacted therapeutics, personalized diagnostics, drug discovery, and medical imaging. Although, in many situations, AI clinical decision-support tools may seem superior to rule-based tools, their use may result in additional challenges. Examples include the paucity of large datasets and the presence of unbalanced data (i.e., due to the low occurrence of adverse outcomes), as often seen in neonatal medicine. The most recent and impactful applications of AI in neonatal medicine are discussed in this review, highlighting future research directions relating to the neonatal population. Current AI applications tested in neonatology include tools for vital signs monitoring, disease prediction (respiratory distress syndrome, bronchopulmonary dysplasia, apnea of prematurity) and risk stratification (retinopathy of prematurity, intestinal perforation, jaundice), neurological diagnostic and prognostic support (electroencephalograms, sleep stage classification, neuroimaging), and novel image recognition technologies, which are particularly useful for prompt recognition of infections. To have these kinds of tools helping neonatologists in daily clinical practice could be something extremely revolutionary in the next future. On the other hand, it is important to recognize the limitations of AI to ensure the proper use of this technology.

Precision diagnostics in children

Medical practice is transforming from a reactive to a pro-active and preventive discipline that is underpinned by precision medicine. The advances in technologies in such fields as genomics, proteomics, metabolomics, transcriptomics and artificial intelligence have resulted in a paradigm shift in our understanding of specific diseases in childhood, greatly enhanced by our ability to combine data from changes within cells to the impact of environmental and population changes. Diseases in children have been reclassified as we understand more about their genomic origin and their evolution. Genomic discoveries, additional 'omics' data and advances such as optical genome mapping have driven rapid improvements in the precision and speed of diagnoses of diseases in children and are now being incorporated into newborn screening, have improved targeted therapies in childhood and have supported the development of predictive biomarkers to assess therapeutic impact and determine prognosis in congenital and acquired diseases of childhood. New medical device technologies are facilitating data capture at a population level to support higher diagnostic accuracy and tailored therapies in children according to predicted population outcome, and digital ecosystems now tailor therapies and provide support for their specific needs. By capturing biological and environmental data as early as possible in childhood, we can understand factors that predict disease or maintain health and track changes across a more extensive longitudinal path. Data from multiple health and external sources over long-time periods starting from birth or even in the in utero environment will provide further clarity about how to sustain health and prevent or predict disease. In this respect, we will not only use data to diagnose disease, but precision diagnostics will aid the 'diagnosis of good health'. The principle of 'start early and change more' will thus underpin the value of applying a personalised medicine approach early in life.

Machine learning for the early prediction of infants with electrographic seizures in neonatal hypoxic-ischemic encephalopathy

OBJECTIVE

To assess if early clinical and electroencephalography (EEG) features predict later seizure development in infants with hypoxic-ischemic encephalopathy (HIE).

METHODS

Clinical and EEG parameters <12 h of birth from infants with HIE across eight European Neonatal Units were used to develop seizure-prediction models. Clinical parameters included intrapartum complications, fetal distress, gestational age, delivery mode, gender, birth weight, Apgar scores, assisted ventilation, cord pH, and blood gases. The earliest EEG hour provided a qualitative analysis (discontinuity, amplitude, asymmetry/asynchrony, sleep-wake cycle [SWC]) and a quantitative analysis (power, discontinuity, spectral distribution, inter-hemispheric connectivity) from full montage and two-channel amplitude-integrated EEG (aEEG). Subgroup analysis, only including infants without anti-seizure medication (ASM) prior to EEG was also performed. Machine-learning (ML) models (random forest and gradient boosting algorithms) were developed to predict infants who would later develop seizures and assessed using Matthews correlation coefficient (MCC) and area under the receiver-operating characteristic curve (AUC).

RESULTS

The study included 162 infants with HIE (53 had seizures). Low Apgar, need for ventilation, high lactate, low base excess, absent SWC, low EEG power, and increased EEG discontinuity were associated with seizures. The following predictive models were developed: clinical (MCC 0.368, AUC 0.681), qualitative EEG (MCC 0.467, AUC 0.729), quantitative EEG (MCC 0.473, AUC 0.730), clinical and qualitative EEG (MCC 0.470, AUC 0.721), and clinical and quantitative EEG (MCC 0.513, AUC 0.746). The clinical and qualitative-EEG model significantly outperformed the clinical model alone (MCC 0.470 vs 0.368, p-value .037). The clinical and quantitative-EEG model significantly outperformed the clinical model (MCC 0.513 vs 0.368, p-value .012). The clinical and quantitative-EEG model for infants without ASM (n = 131) had MCC 0.588, AUC 0.832. Performance for quantitative aEEG (n = 159) was MCC 0.381, AUC 0.696 and clinical and quantitative aEEG was MCC 0.384, AUC 0.720.

SIGNIFICANCE

Early EEG background analysis combined with readily available clinical data helped predict infants who were at highest risk of seizures, hours before they occur. Automated quantitative-EEG analysis was as good as expert analysis for predicting seizures, supporting the use of automated assessment tools for early evaluation of HIE.

Heart rate variability analysis for the prediction of EEG grade in infants with hypoxic ischaemic encephalopathy within the first 12 h of birth

Background and aims

Heart rate variability (HRV) has previously been assessed as a biomarker for brain injury and prognosis in neonates. The aim of this cohort study was to use HRV to predict the electroencephalography (EEG) grade in neonatal hypoxic-ischaemic encephalopathy (HIE) within the first 12 h.

Methods

We included 120 infants with HIE recruited as part of two European multi-centre studies, with electrocardiography (ECG) and EEG monitoring performed before 12 h of age. HRV features and EEG background were assessed using the earliest 1 h epoch of ECG-EEG monitoring. HRV was expressed in time, frequency and complexity features. EEG background was graded from 0-normal, 1-mild, 2-moderate, 3-major abnormalities to 4-inactive. Clinical parameters known within 6 h of birth were collected (intrapartum complications, foetal distress, gestational age, mode of delivery, gender, birth weight, Apgar at 1 and 5, assisted ventilation at 10 min). Using logistic regression analysis, prediction models for EEG severity were developed for HRV features and clinical parameters, separately and combined. Multivariable model analysis included 101 infants without missing data.

Results

Of 120 infants included, 54 (45%) had normal-mild and 66 (55%) had moderate-severe EEG grade. The performance of HRV model was AUROC 0.837 (95% CI: 0.759-0.914) and clinical model was AUROC 0.836 (95% CI: 0.759-0.914). The HRV and clinical model combined had an AUROC of 0.895 (95% CI: 0.832-0.958). Therapeutic hypothermia and anti-seizure medication did not affect the model performance.

Conclusions

Early HRV and clinical information accurately predicted EEG grade in HIE within the first 12 h of birth. This might be beneficial when EEG monitoring is not available in the early postnatal period and for referral centres who may want some objective information on HIE severity.

Why monitor the neonatal brain-that is the important question

A key goal of neonatal neurocritical care is improved outcomes, and brain monitoring plays an essential role. The recent NEST trial reported no outcome benefits using aEEG monitoring compared to clinical seizure identification among neonates treated for seizures. However, the study failed to prove the effects of monitoring on seizure treatment in the first place.

Neuromonitoring in neonatal critical care part I: neonatal encephalopathy and neonates with possible seizures

The blooming of neonatal neurocritical care over the last decade reflects substantial advances in neuromonitoring and neuroprotection. The most commonly used brain monitoring tools in the neonatal intensive care unit (NICU) are amplitude integrated EEG (aEEG), full multichannel continuous EEG (cEEG), and near-infrared spectroscopy (NIRS). While some published guidelines address individual tools, there is no consensus on consistent, efficient, and beneficial use of these modalities in common NICU scenarios. This work reviews current evidence to assist decision making for best utilization of neuromonitoring modalities in neonates with encephalopathy or with possible seizures. Neuromonitoring approaches in extremely premature and critically ill neonates are discussed separately in the companion paper. IMPACT: Neuromonitoring techniques hold promise for improving neonatal care. For neonatal encephalopathy, aEEG can assist in screening for eligibility for therapeutic hypothermia, though should not be used to exclude otherwise eligible neonates. Continuous cEEG, aEEG and NIRS through rewarming can assist in prognostication. For neonates with possible seizures, cEEG is the gold standard for detection and diagnosis. If not available, aEEG as a screening tool is superior to clinical assessment alone. The use of seizure detection algorithms can help with timely seizures detection at the bedside.

An automated bedside measure for monitoring neonatal cortical activity: a supervised deep learning-based electroencephalogram classifier with external cohort validation

BACKGROUND

Electroencephalogram (EEG) monitoring is recommended as routine in newborn neurocritical care to facilitate early therapeutic decisions and outcome predictions. EEG's larger-scale implementation is, however, hindered by the shortage of expertise needed for the interpretation of spontaneous cortical activity, the EEG background. We developed an automated algorithm that transforms EEG recordings to quantified interpretations of EEG background and provides simple intuitive visualisations in patient monitors.

METHODS

In this method-development and proof-of-concept study, we collected visually classified EEGs from infants recovering from birth asphyxia or stroke. We used unsupervised learning methods to explore latent EEG characteristics, which guided the supervised training of a deep learning-based classifier. We assessed the classifier performance using cross-validation and an external validation dataset. We constructed a novel measure of cortical function, brain state of the newborn (BSN), from the novel EEG background classifier and a previously published sleep-state classifier. We estimated clinical utility of the BSN by identification of two key items in newborn brain monitoring, the onset of continuous cortical activity and sleep-wake cycling, compared with the visual interpretation of the raw EEG signal and the amplitude-integrated (aEEG) trend.

FINDINGS

We collected 2561 h of EEG from 39 infants (gestational age 35·0-42·1 weeks; postnatal age 0-7 days). The external validation dataset included 105 h of EEG from 31 full-term infants. The overall accuracy of the EEG background classifier was 92% in the whole cohort (95% CI 91-96; range 85-100 for individual infants). BSN trend values were closely related to the onset of continuous EEG activity or sleep-wake cycling, and BSN levels showed robust difference between aEEG categories. The temporal evolution of the BSN trends showed early diverging trajectories in infants with severely abnormal outcomes.

INTERPRETATION

The BSN trend can be implemented in bedside patient monitors as an EEG interpretation that is intuitive, transparent, and clinically explainable. A quantitative trend measure of brain function might harmonise practices across medical centres, enable wider use of brain monitoring in neurocritical care, and might facilitate clinical intervention trials.

FUNDING

European Training Networks Funding Scheme, the Academy of Finland, Finnish Pediatric Foundation (Lastentautiensäätiö), Aivosäätiö, Sigrid Juselius Foundation, HUS Children's Hospital, HUS Diagnostic Center, National Health and Medical Research Council of Australia.

Sleep State Trend (SST), a bedside measure of neonatal sleep state fluctuations based on single EEG channels

OBJECTIVE

To develop and validate an automated method for bedside monitoring of sleep state fluctuations in neonatal intensive care units.

METHODS

A deep learning-based algorithm was designed and trained using 53 EEG recordings from a long-term (a)EEG monitoring in 30 near-term neonates. The results were validated using an independent dataset from 30 polysomnography recordings. In addition, we constructed Sleep State Trend (SST), a bedside-ready means for visualizing classifier outputs.

RESULTS

The accuracy of quiet sleep detection in the training data was 90%, and the accuracy was comparable (85-86 %) in all bipolar derivations available from the 4-electrode recordings. The algorithm generalized well to a polysomnography dataset, showing 81% overall accuracy despite different signal derivations. SST allowed an intuitive, clear visualization of the classifier output.

CONCLUSIONS

Fluctuations in sleep states can be detected at high fidelity from a single EEG channel, and the results can be visualized as a transparent and intuitive trend in the bedside monitors.

SIGNIFICANCE

The Sleep State Trend (SST) may provide caregivers and clinical studies a real-time view of sleep state fluctuations and its cyclicity.

Neuroprotective therapies in the NICU in term infants: present and future

Outcomes of neonatal encephalopathy (NE) have improved since the widespread implementation of therapeutic hypothermia (TH) in high-resource settings. While TH for NE in term and near-term infants has proven beneficial, 30-50% of infants with moderate-to-severe NE treated with TH still suffer death or significant impairments. There is therefore a critical need to find additional pharmacological and non-pharmacological interventions that improve the outcomes for these children. There are many potential candidates; however, it is unclear whether these interventions have additional benefits when used with TH. Although primary and delayed (secondary) brain injury starting in the latent phase after HI are major contributors to neurodisability, the very late evolving effects of tertiary brain injury likely require different interventions targeting neurorestoration. Clinical trials of seizure management and neuroprotection bundles are needed, in addition to current trials combining erythropoietin, stem cells, and melatonin with TH. IMPACT: The widespread use of therapeutic hypothermia (TH) in the treatment of neonatal encephalopathy (NE) has reduced the associated morbidity and mortality. However, 30-50% of infants with moderate-to-severe NE treated with TH still suffer death or significant impairments. This review details the pathophysiology of NE along with the evidence for the use of TH and other beneficial neuroprotective strategies used in term infants. We also discuss treatment strategies undergoing evaluation at present as potential adjuvant treatments to TH in NE.

Application and potential of artificial intelligence in neonatal medicine

Neonatal care is becoming increasingly complex with large amounts of rich, routinely recorded physiological, diagnostic and outcome data. Artificial intelligence (AI) has the potential to harness this vast quantity and range of information and become a powerful tool to support clinical decision making, personalised care, precise prognostics, and enhance patient safety. Current AI approaches in neonatal medicine include tools for disease prediction and risk stratification, neurological diagnostic support and novel image recognition technologies. Key to the integration of AI in neonatal medicine is the understanding of its limitations and a standardised critical appraisal of AI tools. Barriers and challenges to this include the quality of datasets used, performance assessment, and appropriate external validation and clinical impact studies. Improving digital literacy amongst healthcare professionals and cross-disciplinary collaborations are needed to harness the full potential of AI to help take the next significant steps in improving neonatal outcomes for high-risk infants.

Randomized Clinical Trials of Machine Learning Interventions in Health Care: A Systematic Review

IMPORTANCE

Despite the potential of machine learning to improve multiple aspects of patient care, barriers to clinical adoption remain. Randomized clinical trials (RCTs) are often a prerequisite to large-scale clinical adoption of an intervention, and important questions remain regarding how machine learning interventions are being incorporated into clinical trials in health care.

OBJECTIVE

To systematically examine the design, reporting standards, risk of bias, and inclusivity of RCTs for medical machine learning interventions.

EVIDENCE REVIEW

In this systematic review, the Cochrane Library, Google Scholar, Ovid Embase, Ovid MEDLINE, PubMed, Scopus, and Web of Science Core Collection online databases were searched and citation chasing was done to find relevant articles published from the inception of each database to October 15, 2021. Search terms for machine learning, clinical decision-making, and RCTs were used. Exclusion criteria included implementation of a non-RCT design, absence of original data, and evaluation of nonclinical interventions. Data were extracted from published articles. Trial characteristics, including primary intervention, demographics, adherence to the CONSORT-AI reporting guideline, and Cochrane risk of bias were analyzed.

FINDINGS

Literature search yielded 19 737 articles, of which 41 RCTs involved a median of 294 participants (range, 17-2488 participants). A total of 16 RCTS (39%) were published in 2021, 21 (51%) were conducted at single sites, and 15 (37%) involved endoscopy. No trials adhered to all CONSORT-AI standards. Common reasons for nonadherence were not assessing poor-quality or unavailable input data (38 trials [93%]), not analyzing performance errors (38 [93%]), and not including a statement regarding code or algorithm availability (37 [90%]). Overall risk of bias was high in 7 trials (17%). Of 11 trials (27%) that reported race and ethnicity data, the median proportion of participants from underrepresented minority groups was 21% (range, 0%-51%).

CONCLUSIONS AND RELEVANCE

This systematic review found that despite the large number of medical machine learning-based algorithms in development, few RCTs for these technologies have been conducted. Among published RCTs, there was high variability in adherence to reporting standards and risk of bias and a lack of participants from underrepresented minority groups. These findings merit attention and should be considered in future RCT design and reporting.

Randomized Controlled Trials of Artificial Intelligence in Clinical Practice: Systematic Review

BACKGROUND

The number of artificial intelligence (AI) studies in medicine has exponentially increased recently. However, there is no clear quantification of the clinical benefits of implementing AI-assisted tools in patient care.

OBJECTIVE

This study aims to systematically review all published randomized controlled trials (RCTs) of AI-assisted tools to characterize their performance in clinical practice.

METHODS

CINAHL, Cochrane Central, Embase, MEDLINE, and PubMed were searched to identify relevant RCTs published up to July 2021 and comparing the performance of AI-assisted tools with conventional clinical management without AI assistance. We evaluated the primary end points of each study to determine their clinical relevance. This systematic review was conducted following the updated PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 guidelines.

RESULTS

Among the 11,839 articles retrieved, only 39 (0.33%) RCTs were included. These RCTs were conducted in an approximately equal distribution from North America, Europe, and Asia. AI-assisted tools were implemented in 13 different clinical specialties. Most RCTs were published in the field of gastroenterology, with 15 studies on AI-assisted endoscopy. Most RCTs studied biosignal-based AI-assisted tools, and a minority of RCTs studied AI-assisted tools drawn from clinical data. In 77% (30/39) of the RCTs, AI-assisted interventions outperformed usual clinical care, and clinically relevant outcomes improved with AI-assisted intervention in 70% (21/30) of the studies. Small sample size and single-center design limited the generalizability of these studies.

CONCLUSIONS

There is growing evidence supporting the implementation of AI-assisted tools in daily clinical practice; however, the number of available RCTs is limited and heterogeneous. More RCTs of AI-assisted tools integrated into clinical practice are needed to advance the role of AI in medicine.

TRIAL REGISTRATION

PROSPERO CRD42021286539; https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=286539.

Classification of partial seizures based on functional connectivity: A MEG study with support vector machine

Temporal lobe epilepsy (TLE) is a chronic neurological disorder that is divided into two subtypes, complex partial seizures (CPS) and simple partial seizures (SPS), based on clinical phenotypes. Revealing differences among the functional networks of different types of TLE can lead to a better understanding of the symbology of epilepsy. Whereas Although most studies had focused on differences between epileptic patients and healthy controls, the neural mechanisms behind the differences in clinical representations of CPS and SPS were unclear. In the context of the era of precision, medicine makes precise classification of CPS and SPS, which is crucial. To address the above issues, we aimed to investigate the functional network differences between CPS and SPS by constructing support vector machine (SVM) models. They mainly include magnetoencephalography (MEG) data acquisition and processing, construction of functional connectivity matrix of the brain network, and the use of SVM to identify differences in the resting state functional connectivity (RSFC). The obtained results showed that classification was effective and accuracy could be up to 82.69% (training) and 81.37% (test). The differences in functional connectivity between CPS and SPS were smaller in temporal and insula. The differences between the two groups were concentrated in the parietal, occipital, frontal, and limbic systems. Loss of consciousness and behavioral disturbances in patients with CPS might be caused by abnormal functional connectivity in extratemporal regions produced by post-epileptic discharges. This study not only contributed to the understanding of the cognitive-behavioral comorbidity of epilepsy but also improved the accuracy of epilepsy classification.

Viewpoint: digital paediatrics-so close yet so far away

Technology is driving a revolution in healthcare, but paediatric services have not fully harnessed the potential. Digital health solutions yet to achieve their promise in paediatrics include electronic health records, decision support systems, telemedicine and remote consultations, despite the accelerated uptake during the COVID-19 pandemic. There are also significant potential benefits in digitally enabled research, including systems to identify and recruit participants online or through health records, tools to extract data points from routine data sets rather than new data collection, and remote approaches to outcome measurement. Children and their families are increasingly becoming digital health citizens, able to manage their own health and use of health services through mobile apps and wearables such as fitness trackers. Ironically, one barrier to the uptake of these technologies is that the fast pace of change in this area means the evidence base behind many of these tools remains underdeveloped. Clinicians are often sceptical of innovations which appear largely driven by enthusiasts rather than science. Rigorous studies are needed to demonstrate safety and effectiveness. Regulators need to be agile and responsive. Implementation needs adequate resource and time, and needs to minimise risks and address concerns, such as worries over losing human contact. Digital health care needs to be embedded in medical education and training so that clinicians are trained in the use of innovations and can understand how to embed within services. In this way, digital paediatrics can deliver benefits to the profession, to services and to our patients.