An automated bedside measure for monitoring neonatal cortical activity: a supervised deep learning-based electroencephalogram classifier with external cohort validation.

BACKGROUND: Electroencephalogram (EEG) monitoring is recommended as routine in newborn neurocritical care to facilitate early therapeutic decisions and outcome predictions. EEG's larger-scale implementation is, however, hindered by the shortage of expertise needed for the interpretation of spontaneous cortical activity, the EEG background. We developed an automated algorithm that transforms EEG recordings to quantified interpretations of EEG background and provides simple intuitive visualisations in patient monitors. METHODS: In this method-development and proof-of-concept study, we collected visually classified EEGs from infants recovering from birth asphyxia or stroke. We used unsupervised learning methods to explore latent EEG characteristics, which guided the supervised training of a deep learning-based classifier. We assessed the classifier performance using cross-validation and an external validation dataset. We constructed a novel measure of cortical function, brain state of the newborn (BSN), from the novel EEG background classifier and a previously published sleep-state classifier. We estimated clinical utility of the BSN by identification of two key items in newborn brain monitoring, the onset of continuous cortical activity and sleep-wake cycling, compared with the visual interpretation of the raw EEG signal and the amplitude-integrated (aEEG) trend. FINDINGS: We collected 2561 h of EEG from 39 infants (gestational age 35·0-42·1 weeks; postnatal age 0-7 days). The external validation dataset included 105 h of EEG from 31 full-term infants. The overall accuracy of the EEG background classifier was 92% in the whole cohort (95% CI 91-96; range 85-100 for individual infants). BSN trend values were closely related to the onset of continuous EEG activity or sleep-wake cycling, and BSN levels showed robust difference between aEEG categories. The temporal evolution of the BSN trends showed early diverging trajectories in infants with severely abnormal outcomes. INTERPRETATION: The BSN trend can be implemented in bedside patient monitors as an EEG interpretation that is intuitive, transparent, and clinically explainable. A quantitative trend measure of brain function might harmonise practices across medical centres, enable wider use of brain monitoring in neurocritical care, and might facilitate clinical intervention trials. FUNDING: European Training Networks Funding Scheme, the Academy of Finland, Finnish Pediatric Foundation (Lastentautiensäätiö), Aivosäätiö, Sigrid Juselius Foundation, HUS Children's Hospital, HUS Diagnostic Center, National Health and Medical Research Council of Australia.

Building an Open Source Classifier for the Neonatal EEG Background: A Systematic Feature-Based Approach From Expert Scoring to Clinical Visualization.

Neonatal brain monitoring in the neonatal intensive care units (NICU) requires a continuous review of the spontaneous cortical activity, i.e., the electroencephalograph (EEG) background activity. This needs development of bedside methods for an automated assessment of the EEG background activity. In this paper, we present development of the key components of a neonatal EEG background classifier, starting from the visual background scoring to classifier design, and finally to possible bedside visualization of the classifier results. A dataset with 13,200 5-minute EEG epochs (8-16 channels) from 27 infants with birth asphyxia was used for classifier training after scoring by two independent experts. We tested three classifier designs based on 98 computational features, and their performance was assessed with respect to scoring system, pre- and post-processing of labels and outputs, choice of channels, and visualization in monitor displays. The optimal solution achieved an overall classification accuracy of 97% with a range across subjects of 81-100%. We identified a set of 23 features that make the classifier highly robust to the choice of channels and missing data due to artefact rejection. Our results showed that an automated bedside classifier of EEG background is achievable, and we publish the full classifier algorithm to allow further clinical replication and validation studies.

Real-time prediction of tumor motion using a dynamic neural network.

Radiation dose delivery into the thoracic and abdomen cavities during radiotherapy treatment is a challenging task as respiratory motion leads to the motion of the target tumor. Real-time repositioning of the treatment beam during radiotherapy requires a method for predicting the tumor motion in order to overcome the inherent electro-mechanical latency of the radiotherapy equipment. Thus, besides respiratory motion, system latency also affects the accuracy of dose delivery. To compensate for the latency, a predictor should be employed to anticipate the position of the tumor and give some time to the radiotherapy system for repositioning the radiation beam. This study investigated the ability of spatio-temporal and dynamic neural networks in predicting tumor displacement caused by respiration. Nine different designs of neural networks with 665-ms prediction horizon were examined. The most accurate result was obtained using a dynamic 35-to-3 neural network which resulted in a mean absolute error of 0.54 ± 0.13 and a root mean square error of 0.57 ± 0.20. Moreover, the proposed predictor model is independent of any time-consuming processes such as real-time retraining and real-time baseline shift averaging. The results are comparable or superior with the current literature in terms of prediction accuracy. Graphical Abstract.

Nonlinear analysis and synthesis of video images using deep dynamic bottleneck neural networks for face recognition.

Nonlinear components extracted from deep structures of bottleneck neural networks exhibit a great ability to express input space in a low-dimensional manifold. Sharing and combining the components boost the capability of the neural networks to synthesize and interpolate new and imaginary data. This synthesis is possibly a simple model of imaginations in human brain where the components are expressed in a nonlinear low dimensional manifold. The current paper introduces a novel Dynamic Deep Bottleneck Neural Network to analyze and extract three main features of videos regarding the expression of emotions on the face. These main features are identity, emotion and expression intensity that are laid in three different sub-manifolds of one nonlinear general manifold. The proposed model enjoying the advantages of recurrent networks was used to analyze the sequence and dynamics of information in videos. It is noteworthy to mention that this model also has also the potential to synthesize new videos showing variations of one specific emotion on the face of unknown subjects. Experiments on discrimination and recognition ability of extracted components showed that the proposed model has an average of 97.77% accuracy in recognition of six prominent emotions (Fear, Surprise, Sadness, Anger, Disgust, and Happiness), and 78.17% accuracy in the recognition of intensity. The produced videos revealed variations from neutral to the apex of an emotion on the face of the unfamiliar test subject which is on average 0.8 similar to reference videos in the scale of the SSIM method.