Challenges of Developing Robust AI for Intrapartum Fetal Heart Rate Monitoring.

Background: CTG remains the only non-invasive tool available to the maternity team for continuous monitoring of fetal well-being during labour. Despite widespread use and investment in staff training, difficulty with CTG interpretation continues to be identified as a problem in cases of fetal hypoxia, which often results in permanent brain injury. Given the recent advances in AI, it is hoped that its application to CTG will offer a better, less subjective and more reliable method of CTG interpretation. Objectives: This mini-review examines the literature and discusses the impediments to the success of AI application to CTG thus far. Prior randomised control trials (RCTs) of CTG decision support systems are reviewed from technical and clinical perspectives. A selection of novel engineering approaches, not yet validated in RCTs, are also reviewed. The review presents the key challenges that need to be addressed in order to develop a robust AI tool to identify fetal distress in a timely manner so that appropriate intervention can be made. Results: The decision support systems used in three RCTs were reviewed, summarising the algorithms, the outcomes of the trials and the limitations. Preliminary work suggests that the inclusion of clinical data can improve the performance of AI-assisted CTG. Combined with newer approaches to the classification of traces, this offers promise for rewarding future development.

Altered Expression of Umbilical Cord Blood Levels of miR-181b and Its Downstream Target mUCH-L1 in Infants with Moderate and Severe Neonatal Hypoxic-Ischaemic Encephalopathy.

Hypoxic-ischaemic encephalopathy (HIE) remains one of the leading causes of neurological disability worldwide. No blood biomarker capable of early detection and classification of injury severity in HIE has been identified. This study aimed to investigate the potential of miRNA-181b (miR-181b) and its downstream target, ubiquitin C-terminal hydrolase-L1 (UCH-L1), to predict the severity of HIE. Full-term infants with perinatal asphyxia were recruited at birth and observed for the development of HIE, along with healthy controls. Levels of miR-181b and messenger UCH-L1 (mUCH-L1) in umbilical cord blood were determined using qRT-PCR. In total, 131 infants; 40 control, 50 perinatal asphyxia without HIE (PA) and 41 HIE, recruited across two separate cohorts (discovery and validation) were included in this study. Significant and consistent downregulation of miR-181b was observed in infants with moderate/severe HIE compared to all other groups in both cohorts: discovery 0.25 (0.16-0.32) vs 0.61 (0.26-1.39), p = 0.027 and validation 0.33 (0.15-1.78) vs 1.2 (0.071-2.09), p = 0.035. mUCH-L1 showed increased expression in infants with HIE in both cohorts. The expression ratio of miR-181b to mUCH-L1 was reduced in those infants with moderate/severe HIE in both cohorts: discovery cohort 0.23 (0.06-0.44) vs 1.59 (0.46-2.54), p = 0.01 and validation cohort 0.41 (0.10-0.81) vs 1.38 (0.59-2.56) in all other infants, p = 0.009. We have validated consistent patterns of altered expression in miR-181b/mUCH-L1 in moderate/severe neonatal HIE which may have the potential to guide therapeutic intervention in HIE.

Bag of Patterns for Nearest Neighbour Neonatal EEG Recall.

Clinical neurophysiologists often find it difficult to recall rare EEG patterns despite the fact that this information could be diagnostic and help with treatment intervention. Traditional search methods may take time to retrieve the archived EEGs that could provide the meaning or cause of the specific pattern, which is undesirable as time can be critical for sick neonates. If neurophysiologists had the ability to quickly recall similar patterns, the prior occurrence of the pattern may help make an earlier diagnosis. This paper presents a system that may be used to assist a clinical neurophysiologist in the recall of neonatal EEG patterns. This paper compares two brute force approaches for the task of neonatal EEG recall and looks at the performance accuracy, speed and memory requirements. This system was tested on six different neonatal EEG pattern types with 430 events in total and the results are presented in this paper.

An Approximate Nearest Neighbour System For Neonatal EEG Recall.

Clinical neurophysiologists often find it difficult to recall rare EEG patterns despite the fact that this information could be diagnostic and help with treatment intervention. Traditional search methods may take time to retrieve the archived EEGs that could provide the meaning or cause of the specific pattern which is not acceptable as time can be critical for sick neonates. If neurophysiologists had the ability to quickly recall similar patterns, the prior occurrence of the pattern may help make an earlier diagnosis. This paper presents a system that may be used to assist a clinical neurophysiologist in the recall of neonatal EEG patterns. The proposed system consists of an alignment technique followed by an approximate nearest neighbour search algorithm called locality sensitive hashing. The system was tested on six different neonatal EEG pattern types with 430 events in total and the results are presented in this paper.

Mild hypoxic ischaemic encephalopathy and long term neurodevelopmental outcome - A systematic review.

AIMS: Hypoxic ischaemic encephalopathy (HIE) remains a significant cause of long term neurodisability despite therapeutic hypothermia (TH). Infants with mild HIE, representing 50% of those with HIE, are perceived as low risk and are currently not eligible for TH [1]. This review examines the available evidence of outcome in term infants with mild HIE. METHODS: Medline, Embase and Cochrane Clinical Trials databases were searched in March 2017. Studies with well-defined HIE grading at birth and standardised neurodevelopmental assessment at ≥18 months were included. Abnormal outcome was defined as death, cerebral palsy or standardised neurodevelopmental test score more than 1 standard deviation below the mean. RESULT: Twenty studies were included. Abnormal outcome was reported in 86/341 (25%) of infants. There was insufficient evidence to examine the effect of TH on outcome. CONCLUSION: A significant proportion of infants with mild HIE have abnormal outcome at follow up.

Monitoring cerebral oxygenation of preterm infants using a neonatal specific sensor.

INTRODUCTION: Cerebral oxygenation (rcSO2) monitoring in preterm infants may identify periods of cerebral hypoxia or hyperoxia. We hypothesised that there was a relationship between rcSO2 values and short term outcome in infants of GA < 32weeks. METHODS: RcSO2 values were recorded for the first 48 h of life using an INVOS monitor with a neonatal sensor. The association between cranial ultrasound scan measured brain injury and rcSO2 was assessed. RESULTS: 120 infants were included. Sixty-nine percent (83) of infants had a normal outcome (no IVH, no PVL, and survival at 1 month); less than one-quarter, 22% (26), had low grade IVH 1 or 2 (moderate outcome); and 9% (11) of infants had a severe outcome (IVH ≥ 3, PVL or died before 1 month age). rcSO2 values were lower for infants GA < 28weeks when compared with those GA 28-32, p < 0.001. There was no difference in absolute rcSO2 values between the three outcome groups but a greater degree of cerebral hypoxia was associated with preterm infants who had low grade 1 or 2 IVH. CONCLUSION: Infants of GA < 28 weeks have lower cerebral oxygenation in the first 2 days of life. A greater degree of hypoxia was seen in infants with grade 1 or 2 haemorrhage. Normative ranges need to be gestation specific.

Downstream mRNA Target Analysis in Neonatal Hypoxic-Ischaemic Encephalopathy Identifies Novel Marker of Severe Injury: a Proof of Concept Paper.

Human microRNA miR-374a is downregulated in the umbilical cord blood (UCB) of infants with hypoxic-ischaemic encephalopathy (HIE). The downstream targets of this microRNA (miRNA) are unclear, but one putative target is the activin-A receptor type IIb (ACVR2B). ACVR2B is required for activin-A function and previous reports have shown alterations of activin-A levels in neonatal HIE. Our aim was to investigate the expression of the potential downstream targets of miR-374a, activin-A and ACVR2B, at birth in a cohort of full-term infants with perinatal asphyxia (PA) only, and those with PA who developed clinical and electrographic HIE. UCB was drawn and processed immediately after delivery. Levels of serum activin-A were measured using ELISA. mRNA levels of ACVR2B in whole blood were quantified using qRT-PCR. Outcome was assessed at 3 years of age using standardised developmental assessment. In total, 171 infants were enrolled: 88 healthy controls, 56 PA and 27 HIE. A statistically significant elevation of median (IQR) ACVR2B was detected in infants with severe HIE compared to moderate/mild HIE, PA and control groups (3.3 (2.94-3.67) vs. 0.91 (0.55-1.21) vs. 0.88 (0.57-1.38) vs. 0.84 (0.74-1.24), p values = 0.04, 0.027 and 0.025, respectively). Although serum activin-A levels were elevated in infants with severe HIE, this elevation did not reach significance. ACVR2B may be a potential novel marker of HIE severity. This is the first study to examine the relationship between activin-A, its receptor AVCR2B and potentially upstream miRNA miR-374a in a cohort of carefully categorised and phenotyped infants. We have shown that miRNA analysis, combined with downstream target exploration, may yield novel biomarkers for the prediction of HIE severity.

Early Cord Metabolite Index and Outcome in Perinatal Asphyxia and Hypoxic-Ischaemic Encephalopathy.

BACKGROUND: A 1H-NMR-derived metabolomic index based on early umbilical cord blood alterations of succinate, glycerol, 3-hydroxybutyrate and O-phosphocholine has shown potential for the prediction of hypoxic-ischaemic encephalopathy (HIE) severity. OBJECTIVE: To evaluate whether this metabolite score can predict 3-year neurodevelopmental outcome in infants with perinatal asphyxia and HIE, compared with current standard biochemical and clinical markers. METHODS: From September 2009 to June 2011, infants at risk of perinatal asphyxia were recruited from a single maternity hospital. Cord blood was drawn and biobanked at delivery. Neonates were monitored for development of encephalopathy both clinically and electrographically. Neurodevelopmental outcome was assessed at 36-42 months using the Bayley Scales of Infant and Toddler Development, ed. III (BSID-III). Death and cerebral palsy were also considered as abnormal end points. RESULTS: Thirty-one infants had both metabolomic analysis and neurodevelopmental outcome at 36-42 months. No child had a severely abnormal BSID-III result. The metabolite index significantly correlated with outcome (ρ2 = 0.30, p < 0.01), which is robust to predict both severe outcome (area under the receiver operating characteristic curve: 0.92, p < 0.01) and intact survival (0.80, p = 0.01). There was no correlation between the index score and performance in the individual BSID-III subscales (cognitive, language, motor). CONCLUSIONS: The metabolite index outperformed other standard biochemical markers at birth for prediction of outcome at 3 years, but was not superior to EEG or the Sarnat score.

Estimating functional brain maturity in very and extremely preterm neonates using automated analysis of the electroencephalogram.

OBJECTIVE: To develop an automated estimate of EEG maturational age (EMA) for preterm neonates. METHODS: The EMA estimator was based on the analysis of hourly epochs of EEG from 49 neonates with gestational age (GA) ranging from 23 to 32weeks. Neonates had appropriate EEG for GA based on visual interpretation of the EEG. The EMA estimator used a linear combination (support vector regression) of a subset of 41 features based on amplitude, temporal and spatial characteristics of EEG segments. Estimator performance was measured with the mean square error (MSE), standard deviation of the estimate (SD) and the percentage error (SE) between the known GA and estimated EMA. RESULTS: The EMA estimator provided an unbiased estimate of EMA with a MSE of 82days (SD=9.1days; SE=4.8%) which was significantly lower than a nominal reading (the mean GA in the dataset; MSE of 267days, SD of 16.3days, SE=8.4%: p<0.001). The EMA estimator with the lowest MSE used amplitude, spatial and temporal EEG characteristics. CONCLUSIONS: The proposed automated EMA estimator provides an accurate estimate of EMA in early preterm neonates. SIGNIFICANCE: Automated analysis of the EEG provides a widely accessible, noninvasive and continuous assessment of functional brain maturity.

In-depth performance analysis of an EEG based neonatal seizure detection algorithm.

OBJECTIVE: To describe a novel neurophysiology based performance analysis of automated seizure detection algorithms for neonatal EEG to characterize features of detected and non-detected seizures and causes of false detections to identify areas for algorithmic improvement. METHODS: EEGs of 20 term neonates were recorded (10 seizure, 10 non-seizure). Seizures were annotated by an expert and characterized using a novel set of 10 criteria. ANSeR seizure detection algorithm (SDA) seizure annotations were compared to the expert to derive detected and non-detected seizures at three SDA sensitivity thresholds. Differences in seizure characteristics between groups were compared using univariate and multivariate analysis. False detections were characterized. RESULTS: The expert detected 421 seizures. The SDA at thresholds 0.4, 0.5, 0.6 detected 60%, 54% and 45% of seizures. At all thresholds, multivariate analyses demonstrated that the odds of detecting seizure increased with 4 criteria: seizure amplitude, duration, rhythmicity and number of EEG channels involved at seizure peak. Major causes of false detections included respiration and sweat artefacts or a highly rhythmic background, often during intermediate sleep. CONCLUSION: This rigorous analysis allows estimation of how key seizure features are exploited by SDAs. SIGNIFICANCE: This study resulted in a beta version of ANSeR with significantly improved performance.

Application of audio fingerprinting to neonatal EEG.

A clinical neurophysiologist must recognize patterns in EEG signals to evaluate the health of a patient's brain activity. Rare or unusual patterns may take time to correctly identify. The ability to automatically assist this recall would be beneficial in ensuring that appropriate measures could be taken in a timely fashion. Audio fingerprinting is a method used to identify songs using only a snippet of the song. Fingerprints are extracted from a sub-section of the song and matched against a database of previously computed fingerprints. In this paper, a fingerprint quantization technique is implemented on neonatal EEG data to attempt to identify sections of EEG data when only seeing a sub-section of the data. The impact of signal distortions is investigated and results from a database of one hour recordings from 40 newborns are presented.

The effect of haemolysis on the metabolomic profile of umbilical cord blood.

OBJECTIVES: Metabolomics is defined as the comprehensive study of all low molecular weight biochemicals, (metabolites) present in an organism. Using a systems biology approach, metabolomics in umbilical cord blood (UCB) may offer insight into many perinatal disease processes by uniquely detecting rapid biochemical pathway alterations. In vitro haemolysis is a common technical problem affecting UCB sampling in the delivery room, and can hamper metabolomic analysis. The extent of metabolomic alteration which occurs in haemolysed samples is unknown. DESIGN AND METHODS: Visual haemolysis was designated by the laboratory technician using a standardised haemolysis index colour chart. The metabolomic profile of haemolysed and non-haemolysed UCB serum samples from 69 healthy term infants was compared using both (1)H-NMR and targeted DI and LC-MS/MS approach. RESULTS: We identified 43 metabolites that are significantly altered in visually haemolysed UCB samples, acylcarnitines (n=2), glycerophospholipids (n=23), sphingolipids (n=7), sugars (n=3), amino acids (n=4) and Krebs cycle intermediates (n=4). CONCLUSION: This information will be useful for researchers in the field of neonatal metabolomics to avoid false findings in the presence of haemolysis, to ensure reproducible and credible results.

Stream computing for biomedical signal processing: A QRS complex detection case-study.

Recent developments in "Big Data" have brought significant gains in the ability to process large amounts of data on commodity server hardware. Stream computing is a relatively new paradigm in this area, addressing the need to process data in real time with very low latency. While this approach has been developed for dealing with large scale data from the world of business, security and finance, there is a natural overlap with clinical needs for physiological signal processing. In this work we present a case study of streams processing applied to a typical physiological signal processing problem: QRS detection from ECG data.

Artefact detection in neonatal EEG.

Artefact detection is an important component of any automated EEG analysis. It is of particular importance in analyses such as sleep state detection and EEG grading where there is no null state. We propose a general artefact detection system (GADS) based on the analysis of the neonatal EEG. This system aims to detect both major and minor artefacts (a distinction based primarily on amplitude). As a result, a two-stage system was constructed based on 14 features extracted from EEG epochs at multiple time scales: [2, 4, 16, 32]s. These features were combined in a support vector machine (SVM) in order to determine the presence of absence of artefact. The performance of the GADS was estimated using a leave-one-out cross-validation applied to a database of hour long recordings from 51 neonates. The median AUC was 1.00 (IQR: 0.95-1.00) for the detection of major artefacts and 0.89 (IQR: 0.83-0.95) for the detection of minor artefacts.

Robustness of time frequency distribution based features for automated neonatal EEG seizure detection.

In this paper we examined the robustness of a feature-set based on time-frequency distributions (TFDs) for neonatal EEG seizure detection. This feature-set was originally proposed in literature for neonatal seizure detection using a support vector machine (SVM). We tested the performance of this feature-set with a smoothed Wigner-Ville distribution and modified B distribution as the underlying TFDs. The seizure detection system using time-frequency signal and image processing features from the TFD of the EEG signal using modified B distribution was able to achieve a median receiver operator characteristic area of 0.96 (IQR 0.91-0.98) tested on a large clinical dataset of 826 h of EEG data from 18 full-term newborns with 1389 seizures. The mean AUC was 0.93.

Normative levels of Interleukin 16 in umbilical cord blood.

OBJECTIVES: The need for early and accurate prediction of outcome in hypoxic-ischaemic encephalopathy (HIE) remains critical. We have previously demonstrated that Interleukin 16 (IL-16) is raised in the umbilical cord blood (UCB) of infants with moderate and severe HIE and has the potential to be developed as a predictive biomarker. Normal reference ranges for IL-16 in UCB have not been previously described. The aim of this study was to determine normative levels of IL-16 in full term neonates using UCB following uncomplicated deliveries and to examine the effect of labour on cord IL-16 values. DESIGN AND METHODS: Full term infants were recruited as part of an ongoing birth cohort study, the Cork BASELINE Birth Cohort Study. All had UCB drawn and bio-banked at -80°C, within 3hours of birth. Samples for this experiment were chosen from this population based cohort study to represent uncomplicated pre-labour caesarean sections and spontaneous vaginal deliveries. Analysis was performed on plasma EDTA, using ELISA Quantikine® (R&D Systems, Europe). RESULTS: Samples were analysed from 48 infants with two modes of delivery; spontaneous vaginal delivery (n=12 male, n=12 female) and elective caesarean section (n=12 male, n=12 female). The range of all samples was normally distributed between 87.0 and 114.6pg/ml. Overall mean (SD) for IL-16 was 102.9 (21.5) pg/ml. Levels were not affected by spontaneous vaginal delivery or gender. CONCLUSION: For the first time we have described the expected range of cord plasma IL-16 levels in healthy term infants following pre-labour and post-labour delivery.

An automated system for grading EEG abnormality in term neonates with hypoxic-ischaemic encephalopathy.

Automated analysis of the neonatal EEG has the potential to assist clinical decision making for neonates with hypoxic-ischaemic encephalopathy. This paper proposes a method of automatically grading the degree of abnormality in an hour long epoch of neonatal EEG. The automated grading system (AGS) was based on a multi-class linear classifier grading of short-term epochs of EEG which were converted into a long-term grading of EEG using a majority vote operation. The features used in the AGS were summary measurements of two sub-signals extracted from a quadratic time-frequency distribution: the amplitude modulation and instantaneous frequency. These sub-signals were based on a model of EEG as a multiplication of a coloured random process with a slowly varying pseudo-periodic waveform and may be related to macroscopic neurophysiological function. The 4 grade AGS had a classification accuracy of 83% compared to human annotation of the EEG (level of agreement, κ = 0.76). Features estimated on the developed sub-signals proved more effective at grading the EEG than measures based solely on the EEG and the incorporation of additional sub-grades based on EEG states into the AGS also improved performance.

Association of nucleated red blood cells and severity of encephalopathy in normothermic and hypothermic infants.

AIM: To determine whether hypothermia alters the discriminative ability of postnatal nucleated red blood cells (NRBCs) to distinguish between mild and moderate/severely encephalopathic infants. METHODS: A prospective cohort study recruited full-term neonates with hypoxic ischaemic encephalopathy (HIE) from 2003 to 2012 (prehypothermic and hypothermic eras). The NRBC count was analysed in the first 24 h in all infants and compared between normothermic and hypothermic cohorts. The severity of encephalopathy was categorized using both clinical Sarnat score and continuous multichannel EEG. RESULTS: Eighty-six infants with HIE were included: in the normothermic group, 19 were clinically mild, 24 moderate/severe; in the hypothermic group, 22 were mild, 21 moderate/severe encephalopathy. NRBC count discriminated between mild and moderate/severe Sarnat scores in the normothermic group (p = 0.03) but not in the hypothermic group (p = 0.9). This change was due to a decrease in NRBCs among moderately encephalopathic infants in the hypothermic cohort. CONCLUSION: Postnatal NRBCs distinguished between mild and moderate/severe encephalopathy in normothermic infants but not in infants undergoing therapeutic hypothermia. We advise caution when using postnatal blood samples to study diagnostic biomarkers for HIE without first analysing the potential impact of hypothermia upon these markers.

A nonparametric feature for neonatal EEG seizure detection based on a representation of pseudo-periodicity.

Automated methods of neonatal EEG seizure detection attempt to highlight the evolving, stereotypical, pseudo-periodic, nature of EEG seizure while rejecting the nonstationary, modulated, coloured stochastic background in the presence of various EEG artefacts. An important aspect of neonatal seizure detection is, therefore, the accurate representation and detection of pseudo-periodicity in the neonatal EEG. This paper describes a method of detecting pseudo-periodic components associated with neonatal EEG seizure based on a novel signal representation; the nonstationary frequency marginal (NFM). The NFM can be considered as an alternative time-frequency distribution (TFD) frequency marginal. This method integrates the TFD along data-dependent, time-frequency paths that are automatically extracted from the TFD using an edge linking procedure and has the advantage of reducing the dimension of a TFD. The reduction in dimension simplifies the process of estimating a decision statistic designed for the detection of the pseudo-periodicity associated with neonatal EEG seizure. The use of the NFM resulted in a significant detection improvement compared to existing stationary and nonstationary methods. The decision statistic estimated using the NFM was then combined with a measurement of EEG amplitude and nominal pre- and post-processing stages to form a seizure detection algorithm. This algorithm was tested on a neonatal EEG database of 18 neonates, 826 h in length with 1389 seizures, and achieved comparable performance to existing second generation algorithms (a median receiver operating characteristic area of 0.902; IQR 0.835-0.943 across 18 neonates).

The use of conventional EEG for the assessment of hypoxic ischaemic encephalopathy in the newborn: a review.

Neonatal hypoxic ischaemic encephalopathy continues to be one of the leading causes of morbidity and mortality among neonates around the globe. With the advent of therapeutic hypothermia, the need to accurately classify the severity of injury in the early neonatal period is of great importance. As clinical measures cannot always accurately estimate the severity early enough for treatment to be initiated, clinicians have become more dependent on conventional and amplitude integrated EEG. Despite this, there is currently no single agreed classification scheme for the neonatal EEG in hypoxic ischaemic encephalopathy. In this review we discuss classification schemes of neonatal background EEG, published over the past 35 years, highlighting the urgent need for a universal visual analysis scheme.