2D Wavelet Scalogram Training of Deep Convolutional Neural Network for Automatic Identification of Micro-Scale Sharp Wave Biomarkers in the Hypoxic-Ischemic EEG of Preterm Sheep

Rule-based deep learning method for prognosis of neonatal hypoxic-ischemic encephalopathy by using susceptibility weighted image analysis

OBJECTIVE

Susceptibility weighted imaging (SWI) of neonatal hypoxic-ischemic brain injury can provide assistance in the prognosis of neonatal hypoxic-ischemic encephalopathy (HIE). We propose a convolutional neural network model to classify SWI images with HIE.

MATERIALS AND METHODS

Due to the lack of a large dataset, transfer learning method with fine-tuning a pre-trained ResNet 50 is introduced. We randomly select 11 datasets from patients with normal neurology outcomes (n = 31) and patients with abnormal neurology outcomes (n = 11) at 24 months of age to avoid bias in classification due to any imbalance in the data.

RESULTS

We develop a rule-based system to improve the classification performance, with an accuracy of 0.93 ± 0.09. We also compute heatmaps produced by the Grad-CAM technique to analyze which areas of SWI images contributed more to the classification patients with abnormal neurology outcome.

CONCLUSION

Such regions that are important in the classification accuracy can interpret the relationship between the brain regions affected by hypoxic-ischemic and neurodevelopmental outcomes of infants with HIE at the age of 2 years.

2D Wavelet-Scalogram Deep-Learning for Seizures Pattern Identification in the Post-Hypoxic-Ischemic EEG of Preterm Fetal Sheep

Neonatal seizures after an hypoxic-ischemic (HI) event in preterm newborns can contribute to neural injury and cause impaired brain development. Preterm neonatal seizures are often not detected or their occurrence underestimated. Therefore, there is a need to improve knowledge about preterm seizures that can help establish diagnostic tools for accurate identification of seizures and for determining morphological differences. We have previously shown the superior utility of deep-learning algorithms for the accurate identification and quantification of post-HI microscale epileptiform transients (e.g., gamma spikes and sharp waves) in preterm fetal sheep models; before the irreversible secondary phase of cerebral energy failure starts by the bursts of high-amplitude stereotypic evolving seizures (HAS) in the signal. We have previously developed successful deep-learning algorithms that accurately identify and quantify the micro-scale transients, during the latent phase. Building up on our deep-learning strategies, this work introduces a real-time deep-learning-based pattern fusion approach to identify HAS in the 256Hz sampled post-HI data from our preterm fetuses. Here, for the first time, we propose a 17-layer deep convolutional neural network (CNN) classifier fed with 2D wavelet-scalogram (WS) images of the EEG patterns for accurate seizure identification. The WS-CNN classifier was cross-validated over 1812 manually annotated EEG segments during ~6 to 48 hours post-HI recordings. The classifier accurately recognized HAS patterns with 97.19% overall accuracy (AUC = 0.96).Clinical relevance-The promising results from this preliminary work indicate the ability of the proposed WS-CNN pattern classifier to identify HI-related seizures in the neonatal preterm brain using 256Hz EEG; the frequency commonly used clinically for data collection.

Formation of the pediatric electroretinogram database parameters for the development of doctor’s decisionmaking algorithm

Electroretinography is a non-invasive electrophysiological method standardized by the International Society for Clinical Electrophysiology of Vision (ISCEV). Electroretinography has been used for the clinical application and standardization of electrophysiological protocols for diagnosing the retina since 1989. Electroretinography become fundamental ophthalmological research method that may assesses the state of the retina. To transfer clinical practice to patients the establishment of standardized protocols is an important step. It is important for monitoring successful molecular therapy in retinal degeneration. Retinitis pigmentosa or achromatopsia and, consequently, affected cones or rods photoreceptors is corresponded to complete absent of electrical response. Thus, detection of even modest improvements after therapeutic treatment is required. Standardized protocols allow the implementation of electroretinography under conditions of optimization of sensitivity and specificity during clinical trials. It should be noted that the literature on retinal diseases demonstrates clinical cases in which patients may have several retinal diseases at the same time. In such cases, it is necessary to detect a group of characteristics of electrophysiological signals with high accuracy to improve the application of various diagnostic solutions. The classification of electroretinogram signals depends on the quality of labeled biomedical information or databases, in addition to this, the accuracy of the classification results obtained depends not only on computer technology, but also on the quality of the input data. To date, the analysis of electroretinogram signals is realized manually and largely depends on the experience of clinicians. The development of automated algorithms for analyzing electroretinogram signals may simplify routine processes and improve the quality of diagnosing eye diseases. This article describes the formation of the parameters of pediatric electroretinogram database parameters for the development of doctor’s decision-making algorithm. The signal parameters were obtained by extracting the parameters from the wavelet scalogram of the electroretinogram signal using digital image processing and machine learning methods.

Deep Convolutional Neural Networks for the Accurate Identification of High-Amplitude Stereotypic Epileptiform Seizures in the Post-Hypoxic-Ischemic EEG of Preterm Fetal Sheep

Neonatal seizures after birth may contribute to brain injury after an hypoxic-ischemic (HI) event, impaired brain development and a later life risk for epilepsy. Despite neural immaturity, seizures can also occur in preterm infants. However, surprisingly little is known about their evolution after an HI insult or patterns of expression. An improved understanding of preterm seizures will help facilitate diagnosis and prognosis and the implementation of treatments. This requires improved detection of seizures, including electrographic seizures. We have established a stable preterm fetal sheep model of HI that results in different types of post-HI seizures. These including the expression of epileptiform transients during the latent phase (0-6 h) of cerebral energy recovery, and bursts of high amplitude stereotypic evolving seizures (HAS) during the secondary phase of cerebral energy failure (∼6-72 h). We have previously developed successful automated machine-learning strategies for accurate identification and quantification of the evolving micro-scale EEG patterns (e.g. gamma spikes and sharp waves), during the latent phase. The current paper introduces, for the first time, a real-time approach that employs a 15-layer deep convolutional neural network (CNN) classifier, directly fed with the raw EEG time-series, to identify HAS in the 1024Hz and 256Hz down-sampled data in our preterm fetuses post-HI. The classifier was trained and tested using EEG segments during ∼6 to 48 hours post-HI recordings. The classifier accurately identified HAS with 98.52% accuracy in the 1024Hz and 97.78% in the 256Hz data. Clinical relevance-Results highlight the promising ability of the proposed CNN classifier for accurate identification of HI related seizures in the neonatal preterm brain, if further applied to the current 256Hz clinical recordings, in real-world.

Wavelet Spectral Time-Frequency Training of Deep Convolutional Neural Networks for Accurate Identification of Micro-Scale Sharp Wave Biomarkers in the Post-Hypoxic-Ischemic EEG of Preterm Sheep

Neonatal hypoxic-ischemic encephalopathy (HIE) evolves over different phases of time during recovery. Some neuroprotection treatments are only effective for specific, short windows of time during this evolution of injury. Clinically, we often do not know when an insult may have started, and thus which phase of injury the brain may be experiencing. To improve diagnosis, prognosis and treatment efficacy, we need to establish biomarkers which denote phases of injury. Our pre-clinical research, using preterm fetal sheep, show that micro-scale EEG patterns (e.g. spikes and sharp waves), superimposed on suppressed EEG background, primarily occur during the early recovery from an HI insult (0-6 h), and that numbers of events within the first 2 h are strongly predictive of neural survival. Thus, real-time automated algorithms that could reliably identify EEG patterns in this phase will help clinicians to determine the phases of injury, to help guide treatment options. We have previously developed successful automated machine learning approaches for accurate identification and quantification of HI micro-scale EEG patterns in preterm fetal sheep post-HI. This paper introduces, for the first time, a novel online fusion strategy that employs a high-level wavelet-Fourier (WF) spectral feature extraction method in conjunction with a deep convolutional neural network (CNN) classifier for accurate identification of micro-scale preterm fetal sheep post-HI sharp waves in 1024Hz EEG recordings, along with 256Hz down-sampled data. The classifier was trained and tested over 4120 EEG segments within the first 2 hours latent phase recordings. The WF-CNN classifier can robustly identify sharp waves with considerable high-performance of 99.86% in 1024Hz and 99.5% in 256Hz data. The method is an alternative deep-structure approach with competitive high-accuracy compared to our computationally-intensive WS-CNN sharp wave classifier.

Wavelet Spectral Deep-training of Convolutional Neural Networks for Accurate Identification of High-Frequency Micro-Scale Spike Transients in the Post-Hypoxic-Ischemic EEG of Preterm Sheep

Early diagnosis and prognosis of babies with signs of hypoxic-ischemic encephalopathy (HIE) is currently limited and requires reliable prognostic biomarkers to identify at risk infants. Using our pre-clinical fetal sheep models, we have demonstrated that micro-scale patterns evolve over a profoundly suppressed EEG background within the first 6 hours of recovery, post HI insult. In particular, we have shown that high-frequency micro-scale spike transients (in the gamma frequency band, 80-120Hz) emerge immediately after an HI event, with much higher numbers around 2-2.5 h of the insult, with numbers gradually declining thereafter. We have also shown that the automatically quantified sharp waves in this phase are predictive of neural outcome. Initiation of some neuroprotective treatments within this limited window of opportunity, such as therapeutic hypothermia, optimally reduces neural injury. In clinical practice, it is hard to determine the exact timing of the injury, therefore, reliable automatic identification of EEG transients could be beneficial to help specify the phases of injury. Our team has previously developed successful machine- and deep-learning strategies for the identification of post-HI EEG patterns in an HI preterm fetal sheep model.This paper introduces, for the first time, a novel online fusion approach to train an 11-layers deep convolutional neural network (CNN) classifier using Wavelet-Fourier (WF) spectral features of EEG segments for accurate identification of high-frequency micro-scale spike transients in 1024Hz EEG recordings in our preterm fetal sheep. Sets of robust features were extracted using reverse biorthogonal wavelet (rbio2.8 at scale 7) and considering an 80-120Hz spectral frequency range. The WF-CNN classifier was able to accurately identify spike transients with a reliable high-performance of 99.03±0.86%.Clinical relevance-Results confirm the expertise of the method for the identification of similar patterns in the EEG of neonates in the early hours after birth.

Deep Convolutional Neural Network and Reverse Biorthogonal Wavelet Scalograms for Automatic Identification of High Frequency Micro-Scale Spike Transients in the Post-Hypoxic-Ischemic EEG

Diagnosis of hypoxic-ischemic encephalopathy (HIE) is currently limited and prognostic biological markers are required for early identification of at risk infants at birth. Using pre-clinical data from our fetal sheep models, we have shown that micro-scale EEG patterns, such as high-frequency spikes and sharp waves, evolve superimposed on a significantly suppressed background during the early hours of recovery (0-6 h), after an HI insult. In particular, we have demonstrated that the number of micro-scale gamma spike transients peaks within the first 2-2.5 hours of the insult and automatically quantified sharp waves in this period are predictive of neural outcome. This period of time is optimal for the initiation of neuroprotection treatments such as therapeutic hypothermia, which has a limited window of opportunity for implementation of 6 h or less after an HI insult. Clinically, it is hard to determine when an insult has started and thus the window of opportunity for treatment. Thus, reliable automatic algorithms that could accurately identify EEG patterns that denote the phase of injury is a valuable clinical tool. We have previously developed successful machine-learning strategies for the identification of HI micro-scale EEG patterns in a preterm fetal sheep model of HI. This paper employs, for the first time, reverse biorthogonal Wavelet-Scalograms (WS) as the inputs to a 17-layer deep-trained convolutional neural network (CNN) for the precise identification of high-frequency micro-scale spike transients that occur in the 80-120Hz gamma band during first 2 h period of an HI insult. The rbio-WS-CNN classifier robustly identified spike transients with an exceptionally high-performance of 99.82%.Clinical relevance-The suggested classifier would effectively identify and quantify EEG patterns of a similar morphology in preterm newborns during recovery from an HI-insult.

Potential Prognostic Markers in the Heart Rate Variability Features for Early Diagnosis of Sepsis in the Pediatric Intensive Care Unit using Convolutional Neural Network Classifiers

Blood infection due to different circumstances could immediately develop to an extreme body reaction that leads to a serious life-threatening condition, called Sepsis. Currently, therapeutic protocols through timely antibiotic resuscitation strategies play an important role to fight against the adverse conditions and improve survival. Therefore, timing, and more specifically early diagnosis of the illness, is crucially important for an effective treatment. Studies have indicated that vital signals such as heart rate variability (HRV) could provide potential prognostic biological markers that can help with early detection of sepsis before it is clinically diagnosed through its actual symptoms. Therefore, this study employs neonatal and pediatric electrocardiogram (ECG) to extract 52 hourly sets of linear and non-linear features from the HRV, starting from 24 hours prior to the clinical diagnosis of sepsis in patients with positive blood cultures (n=14). Similar sets of features were also obtained from a non-sepsis control group to create an evaluation benchmark (n=14).In particular, this study initially demonstrates how the variations within the 24 hours values of specific HRV featuresets could effectively reveal prognostic information about the evolution of sepsis, prior to the actual clinical diagnosis. Moreover, this study demonstrates that differences in the values of a particular set of features at 22 hours before the actual clinical diagnosis/symptoms can be reliably used to train a convolutional neural network for automatic classification between the individuals in the sepsis and non-sepsis groups with 88.89±7.86% accuracy.Clinical relevance- Results suggest potential early diagnosis of sepsis through real-time automatic classification of HRV features as prognostic indicators in clinical ECG recordings.

Potential Prognostic Markers in the Heart Rate Variability Features for Early Diagnosis of Sepsis in the Pediatric Intensive Care Unit using Convolutional Neural Network Classifiers

Blood infection due to different circumstances could immediately develop to an extreme body reaction that leads to a serious life-threatening condition, called Sepsis. Currently, therapeutic protocols through timely antibiotic resuscitation strategies play an important role to fight against the adverse conditions and improve survival. Therefore, timing, and more specifically early diagnosis of the illness, is crucially important for an effective treatment. Studies have indicated that vital signals such as heart rate variability (HRV) could provide potential prognostic biological markers that can help with early detection of sepsis before it is clinically diagnosed through its actual symptoms. Therefore, this study employs neonatal and pediatric electrocardiogram (ECG) to extract 52 hourly sets of linear and non-linear features from the HRV, starting from 24 hours prior to the clinical diagnosis of sepsis in patients with positive blood cultures (n=14). Similar sets of features were also obtained from a non-sepsis control group to create an evaluation benchmark (n=14).In particular, this study initially demonstrates how the variations within the 24 hours values of specific HRV feature-sets could effectively reveal prognostic information about the evolution of sepsis, prior to the actual clinical diagnosis. Moreover, this study demonstrates that differences in the values of a particular set of features at 22 hours before the actual clinical diagnosis/symptoms can be reliably used to train a convolutional neural network for automatic classification between the individuals in the sepsis and non-sepsis groups with 88.89±7.86% accuracy.

Automatically Identified Micro-scale Sharp-wave Transients in the Early-Latent Phase of Hypoxic-Ischemic EEG from Preterm Fetal Sheep Reveal Timing Relationship to Subcortical Neuronal Survival

Perinatal Hypoxic-Ischemia Encephalopathy (HIE) in newborn infants, due to birth-related circumstances such as oxygen deprivation in brain cells, is caused by the disruption in blood flow through the umbilical cord. Subcortical neuronal loss due to the HIE can lead to cerebral palsy and other chronic neurological conditions. Pre-clinical EEG studies using in utero sheep have demonstrated that particular micro-scale HI transients emerge along a suppressed EEG background during a latent phase of 3-6 hours, after a severe HI insult. Whilst the nature of these micro-scale transients is not well understood, it has been hypothesized that such transients may be signatures of the evolving hypoxic-ischemic brain injury, possessing the potential to be served as the diagnosis biomarkers for the injury. Cerebral hypothermia is optimally neuroprotective only if administered within the first 2-3 hours post HI insult. Using data from a cohort of in utero preterm fetal sheep (n=5, at 0.7 of gestational age), this paper indicates how the number of automatically quantified micro-scale sharp wave transients from asphyxiated preterm fetal sheep, statistically correlate to the amount of NeuN-positive neurons measured in caudate nucleus of striatum. Different temporal window sizes of 2hrs, 1hr, ½hr and 10mins within the early phase of the latent phase are examined using our developed Wavelet Type-2 Fuzzy classifier for sharp detection. Analyses were narrowed down to 10min intervals to assess where exactly in time the occurrence of the HI micro-scale sharp waves demonstrate a significant correlation. Signal processing wise, results from the sub-windows indicate a timing trend that highlights a positive correlation, between the number of automatic quantifications and the amount of surviving neurons in the preterm brain, permitting the possibility of a point of care (POC) intervention to stop the spread of injury before it becomes irreversible.

Latent Phase Identification of High-Frequency Micro-Scale Gamma Spike Transients in the Hypoxic Ischemic EEG of Preterm Fetal Sheep Using Spectral Analysis and Fuzzy Classifiers

Premature babies are at high risk of serious neurodevelopmental disabilities, which in many cases are related to perinatal hypoxic–ischemic encephalopathy (HIE). Studies of neuroprotection in animal models consistently suggest that treatment must be started as early as possible in the first 6 h after hypoxia–ischemia (HI), the so-called latent phase before secondary deterioration, to improve outcomes. We have shown in preterm sheep that EEG biomarkers of injury, in the form of high-frequency micro-scale spike transients, develop and evolve in this critical latent phase after severe asphyxia. Real-time automatic identification of such events is important for the early and accurate detection of HI injury, so that the right treatment can be implemented at the right time. We have previously reported successful strategies for accurate identification of EEG patterns after HI. In this study, we report an alternative high-performance approach based on the fusion of spectral Fourier analysis and Type-I fuzzy classifiers (FFT-Type-I-FLC). We assessed its performance in over 2520 min of latent phase EEG recordings from seven asphyxiated in utero preterm fetal sheep exposed to a range of different occlusion periods. The FFT-Type-I-FLC classifier demonstrated 98.9 ± 1.0% accuracy for identification of high-frequency spike transients in the gamma frequency band (namely 80–120 Hz) post-HI. The spectral-based approach (FFT-Type-I-FLC classifier) has similar accuracy to our previous reverse biorthogonal wavelets rbio2.8 basis function and type-1 fuzzy classifier (rbio-WT-Type-1-FLC), providing competitive performance (within the margin of error: 0.89%), but it is computationally simpler and would be readily adapted to identify other potentially relevant EEG waveforms.

Applications of advanced signal processing and machine learning in the neonatal hypoxic-ischemic electroencephalogram

Perinatal hypoxic-ischemic-encephalopathy significantly contributes to neonatal death and life-long disability such as cerebral palsy. Advances in signal processing and machine learning have provided the research community with an opportunity to develop automated real-time identification techniques to detect the signs of hypoxic-ischemic-encephalopathy in larger electroencephalography/amplitude-integrated electroencephalography data sets more easily. This review details the recent achievements, performed by a number of prominent research groups across the world, in the automatic identification and classification of hypoxic-ischemic epileptiform neonatal seizures using advanced signal processing and machine learning techniques. This review also addresses the clinical challenges that current automated techniques face in order to be fully utilized by clinicians, and highlights the importance of upgrading the current clinical bedside sampling frequencies to higher sampling rates in order to provide better hypoxic-ischemic biomarker detection frameworks. Additionally, the article highlights that current clinical automated epileptiform detection strategies for human neonates have been only concerned with seizure detection after the therapeutic latent phase of injury. Whereas recent animal studies have demonstrated that the latent phase of opportunity is critically important for early diagnosis of hypoxic-ischemic-encephalopathy electroencephalography biomarkers and although difficult, detection strategies could utilize biomarkers in the latent phase to also predict the onset of future seizures.

Electroencephalogram studies of hypoxic ischemia in fetal and neonatal animal models

Alongside clinical achievements, experiments conducted on animal models (including primate or non-primate) have been effective in the understanding of various pathophysiological aspects of perinatal hypoxic/ischemic encephalopathy (HIE). Due to the reasonably fair degree of flexibility with experiments, most of the research around HIE in the literature has been largely concerned with the neurodevelopmental outcome or how the frequency and duration of HI seizures could relate to the severity of perinatal brain injury, following HI insult. This survey concentrates on how EEG experimental studies using asphyxiated animal models (in rodents, piglets, sheep and non-human primate monkeys) provide a unique opportunity to examine from the exact time of HI event to help gain insights into HIE where human studies become difficult.