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

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.

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.

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.