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

Transformer-Based Wavelet-Scalogram Deep Learning for Improved Seizure Pattern Recognition in Post-Hypoxic-Ischemic Fetal Sheep EEG

Hypoxic-ischemic (HI) events in newborns can trigger seizures, which are highly associated with later neurodevelopmental impairment. The precise detection of these seizures is a complex task requiring considerable very specialized expertise, underscoring the necessity for automated methods to support diagnosis and therapeutic interventions. We have previously shown the effectiveness of deep-learning algorithms, involving a 17-layer deep convolutional neural network (CNN), to identify and quantify post-HI high-amplitude seizures (HAS) in preterm fetal sheep models. This method, which incorporated 2D wavelet scalogram (WS) images of EEG patterns, achieved a cross-validated overall accuracy of 97.19% (AUC=0.96). In the current study, we now assess the utility of Transformer models for seizure detection in the EEG of pre- and near-term fetal sheep during the first 48 hours of recovery after HI. We trained a series of Transformer models on a smaller subset of the original dataset, comprising 800 WS of EEG seizure and non-seizure patterns. Out of all classifiers, the Visual Transformer (ViT) exceeded the performance of the previous deep CNNs, achieving an overall accuracy of 99.5% with an AUC of 0.995.Clinical relevance-This study showcases the superior performance, efficiency, and robustness of Transformer models for identifying HI-induced EEG seizures, with high potential to be clinical decision support tools to identify neonatal seizures at the cotside.

Deep Learning for Generalized EEG Seizure Detection after Hypoxia-Ischemia-Preclinical Validation

Brain maturity and many clinical treatments such as therapeutic hypothermia (TH) can significantly influence the morphology of neonatal EEG seizures after hypoxia-ischemia (HI), and so there is a need for generalized automatic seizure identification. This study validates efficacy of advanced deep-learning pattern classifiers based on a convolutional neural network (CNN) for seizure detection after HI in fetal sheep and determines the effects of maturation and brain cooling on their accuracy. The cohorts included HI-normothermia term (n = 7), HI-hypothermia term (n = 14), sham-normothermia term (n = 5), and HI-normothermia preterm (n = 14) groups, with a total of >17,300 h of recordings. Algorithms were trained and tested using leave-one-out cross-validation and k-fold cross-validation approaches. The accuracy of the term-trained seizure detectors was consistently excellent for HI-normothermia preterm data (accuracy = 99.5%, area under curve (AUC) = 99.2%). Conversely, when the HI-normothermia preterm data were used in training, the performance on HI-normothermia term and HI-hypothermia term data fell (accuracy = 98.6%, AUC = 96.5% and accuracy = 96.9%, AUC = 89.6%, respectively). Findings suggest that HI-normothermia preterm seizures do not contain all the spectral features seen at term. Nevertheless, an average 5-fold cross-validated accuracy of 99.7% (AUC = 99.4%) was achieved from all seizure detectors. This significant advancement highlights the reliability of the proposed deep-learning algorithms in identifying clinically translatable post-HI stereotypic seizures in 256Hz recordings, regardless of maturity and with minimal impact from hypothermia.

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.

Deep-Learning Markerless Tracking of Infant General Movements using Standard Video Recordings

Monitoring spontaneous General Movements (GM) of infants 6-20 weeks post-term age is a reliable tool to assess the quality of neurodevelopment in early infancy. Abnormal or absent GMs are reliable prognostic indicators of whether an infant is at risk of developing neurological impairments and disorders such as cerebral palsy (CP). Therapeutic interventions are most effective at improving neuromuscular outcomes if administered in early infancy. Current clinical protocols require trained assessors to rate videos of infant movements, a time-intensive task. This work proposes a simple, inexpensive, and broadly applicable markerless pose-estimation approach for automatic infant movement tracking using conventional video recordings from handheld devices (e.g., tablets and mobile phones). We leverage the enhanced capabilities of deep-learning technology in image processing to identify 12 anatomical locations (3 per limb) in each video frame, tracking a baby's natural movement throughout the recordings. We validate the capability of resnet152 and a mobile-net-v2-1 to identify body-parts in unseen frames from a full-term male infant, using a novel automatic unsupervised approach that fuses likelihood outputs of a Kalman filter and the deep-nets. Both deep-net models were found to perform very well in the identification of anatomical locations in the unseen data with high average Percentage of Correct Keypoints (aPCK) performances of >99.65% across all locations.Clinical relevance-Results of this research confirm the feasibility of a low-cost and publicly accessible technology to automatically track infants' GMs and diagnose those at higher risk of developing neurological conditions early, when clinical interventions are most effective.

Current and Future Uses of Continuous EEG in the NICU

Continuous EEG (cEEG) is a fundamental neurodiagnostic tool in the care of critically ill neonates and is increasingly recommended. cEEG enhances prognostication via assessment of the background brain activity, plays a role in predicting which neonates are at risk for seizures when combined with clinical factors, and allows for accurate diagnosis and management of neonatal seizures. Continuous EEG is the gold standard method for diagnosis of neonatal seizures and should be used for detection of seizures in high-risk clinical conditions, differential diagnosis of paroxysmal events, and assessment of response to treatment. High costs associated with cEEG are a limiting factor in its widespread implementation. Centralized remote cEEG interpretation, automated seizure detection, and pre-natal EEG are potential future applications of this neurodiagnostic tool.