A Deep Shared Multi-Scale Inception Network Enables Accurate Neonatal Quiet Sleep Detection With Limited EEG Channels.

In this paper, we introduce a new variation of the Convolutional Neural Network Inception block, called Sinc, for sleep stage classification in premature newborn babies using electroencephalogram (EEG). In practice, there are many medical centres where only a limited number of EEG channels are recorded. Existing automated algorithms mainly use multi-channel EEGs which perform poorly when fewer numbers of channels are available. The proposed Sinc utilizes multi-scale analysis to place emphasis on the temporal EEG information to be less dependent on the number of EEG channels. In Sinc, we increase the receptive fields through Inception while by additionally sharing the filters that have similar receptive fields, overfitting is controlled and the number of trainable parameters dramatically reduced. To train and test this model, 96 longitudinal EEG recordings from 26 premature infants are used. The Sinc-based model significantly outperforms state-of-the-art neonatal quiet sleep detection algorithms, with mean Kappa 0.77 ± 0.01 (with 8-channel EEG) and 0.75 ± 0.01 (with a single bipolar channel EEG). This is the first study using Inception-based networks for EEG analysis that utilizes filter sharing to improve efficiency and trainability. The suggested network can successfully detect quiet sleep stages with even a single EEG channel making it more practical especially in the hospital setting where cerebral function monitoring is predominantly used.

A convolutional neural network outperforming state-of-the-art sleep staging algorithms for both preterm and term infants.

OBJECTIVE: To classify sleep states using electroencephalogram (EEG) that reliably works over a wide range of preterm ages, as well as term age. APPROACH: A convolutional neural network is developed to perform 2- and 4-class sleep classification in neonates. The network takes as input an 8-channel 30 s EEG segment and outputs the sleep state probabilities. Apart from simple downsampling of the input and smoothing of the output, the suggested network is an end-to-end algorithm that avoids the need for hand-crafted feature selection or complex pre/post processing steps. To train and test this method, 113 EEG recordings from 42 infants are used. MAIN RESULTS: For quiet sleep detection (the 2-class problem), mean kappa between the network estimate and the ground truth annotated by EEG human experts is 0.76. The sensitivity and specificity are 90% and 88%, respectively. For 4-class classification, mean kappa is 0.64. The averaged sensitivity and specificity (1 versus all) respectively equal 72% and 91%. The results outperform current state-of-the-art methods for which kappa ranges from 0.66 to 0.70 in preterm and from 0.51 to 0.61 in term infants, based on training and testing using the same database. SIGNIFICANCE: The proposed method has the highest reported accuracy for EEG sleep state classification for both preterm and term age neonates.

Neonatal Seizure Detection Using Deep Convolutional Neural Networks.

Identifying a core set of features is one of the most important steps in the development of an automated seizure detector. In most of the published studies describing features and seizure classifiers, the features were hand-engineered, which may not be optimal. The main goal of the present paper is using deep convolutional neural networks (CNNs) and random forest to automatically optimize feature selection and classification. The input of the proposed classifier is raw multi-channel EEG and the output is the class label: seizure/nonseizure. By training this network, the required features are optimized, while fitting a nonlinear classifier on the features. After training the network with EEG recordings of 26 neonates, five end layers performing the classification were replaced with a random forest classifier in order to improve the performance. This resulted in a false alarm rate of 0.9 per hour and seizure detection rate of 77% using a test set of EEG recordings of 22 neonates that also included dubious seizures. The newly proposed CNN classifier outperformed three data-driven feature-based approaches and performed similar to a previously developed heuristic method.

Automated EEG background analysis to identify neonates with hypoxic-ischemic encephalopathy treated with hypothermia at risk for adverse outcome: A pilot study.

BACKGROUND: To improve the objective assessment of continuous video-EEG (cEEG) monitoring of neonatal brain function, the aim was to relate automated derived amplitude and duration parameters of the suppressed periods in the EEG background (dynamic Interburst Interval= dIBIs) after neonatal hypoxic-ischemic encephalopathy (HIE) to favourable or adverse neurodevelopmental outcome. METHODS: Nineteen neonates (gestational age 36-41 weeks) with HIE underwent therapeutic hypothermia and had cEEG-monitoring. EEGs were retrospectively analyzed with a previously developed algorithm to detect the dynamic Interburst Intervals. Median duration and amplitude of the dIBIs were calculated at 1 h-intervals. Sensitivity and specificity of automated EEG background grading for favorable and adverse outcomes were assessed at 6 h-intervals. RESULTS: Dynamic IBI values reached the best prognostic value between 18 and 24 h (AUC of 0.93). EEGs with dIBI amplitude ≥15 µV and duration <10 s had a specificity of 100% at 6-12 h for favorable outcome but decreased subsequently to 67% at 25-42 h. Suppressed EEGs with dIBI amplitude <15 µV and duration >10 s were specific for adverse outcome (89-100%) at 18-24 h (n = 10). Extremely low voltage and invariant EEG patterns were indicative of adverse outcome at all time points. CONCLUSIONS: Automated analysis of the suppressed periods in EEG of neonates with HIE undergoing TH provides objective and early prognostic information. This objective tool can be used in a multimodal strategy for outcome assessment. Implementation of this method can facilitate clinical practice, improve risk stratification and aid therapeutic decision-making. A multicenter trial with a quantifiable outcome measure is warranted to confirm the predictive value of this method in a more heterogeneous dataset.

Interrater agreement in visual scoring of neonatal seizures based on majority voting on a web-based system: The Neoguard EEG database.

OBJECTIVE: To assess interrater agreement based on majority voting in visual scoring of neonatal seizures. METHODS: An online platform was designed based on a multicentre seizure EEG-database. Consensus decision based on 'majority voting' and interrater agreement was estimated using Fleiss' Kappa. The influences of different factors on agreement were determined. RESULTS: 1919 Events extracted from 280h EEG of 71 neonates were reviewed by 4 raters. Majority voting was applied to assign a seizure/non-seizure classification. 44% of events were classified with high, 36% with moderate, and 20% with poor agreement, resulting in a Kappa value of 0.39. 68% of events were labelled as seizures, and in 46%, all raters were convinced about electrographic seizures. The most common seizure duration was <30s. Raters agreed best for seizures lasting 60-120s. There was a significant difference in electrographic characteristics of seizures versus dubious events, with seizures having longer duration, higher power and amplitude. CONCLUSIONS: There is a wide variability in identifying rhythmic ictal and non-ictal EEG events, and only the most robust ictal patterns are consistently agreed upon. Database composition and electrographic characteristics are important factors that influence interrater agreement. SIGNIFICANCE: The use of well-described databases and input of different experts will improve neonatal EEG interpretation and help to develop uniform seizure definitions, useful for evidence-based studies of seizure recognition and management.

Objective differentiation of neonatal EEG background grades using detrended fluctuation analysis.

A quantitative and objective assessment of background electroencephalograph (EEG) in sick neonates remains an everyday clinical challenge. We studied whether long range temporal correlations quantified by detrended fluctuation analysis (DFA) could be used in the neonatal EEG to distinguish different grades of abnormality in the background EEG activity. Long-term EEG records of 34 neonates were collected after perinatal asphyxia, and their background was scored in 1 h epochs (8 h in each neonate) as mild, moderate or severe. We applied DFA on 15 min long, non-overlapping EEG epochs (n = 1088) filtered from 3 to 8 Hz. Our formal feasibility study suggested that DFA exponent can be reliably assessed in only part of the EEG epochs, and in only relatively short time scales (10-60 s), while it becomes ambiguous if longer time scales are considered. This prompted further exploration whether paradigm used for quantifying multifractal DFA (MF-DFA) could be applied in a more efficient way, and whether metrics from MF-DFA paradigm could yield useful benchmark with existing clinical EEG gradings. Comparison of MF-DFA metrics showed a significant difference between three visually assessed background EEG grades. MF-DFA parameters were also significantly correlated to interburst intervals quantified with our previously developed automated detector. Finally, we piloted a monitoring application of MF-DFA metrics and showed their evolution during patient recovery from asphyxia. Our exploratory study showed that neonatal EEG can be quantified using multifractal metrics, which might offer a suitable parameter to quantify the grade of EEG background, or to monitor changes in brain state that take place during long-term brain monitoring.