Applications of deep learning in neonatal brain monitoring
Representative and Author: Dr. Amir H. Ansari
Organization: ESAT - STADIUS, Katholieke Universiteit Leuven
E-mail address:
After delivery, sick newborn babies, which need intensive medical attention, will be admitted to neonatal intensive care units (NICUs) for extra attention and special monitoring. In most NICUs, dozens of physiological vital signals including the electrocardiogram (ECG), heart rate, blood pressure, blood oxygen saturation, are often continuously being monitored. However, many NICUs are suffering from the lack of an appropriate continuous brain monitoring system, although its importance has been proved in literature for decades. The main challenge of the brain monitoring is the analysis of the brain signal, Electroencephalogram (EEG), which needs special expertise that may not be available around the clock. Therefore, a continuous brain monitor equipped with some supportive tools and automated algorithms (e.g. for seizure detection, sleep stage classification, EEG background assessment), is needed in order to decrease the expenses of monitoring and improve the quality of treatment.
In recent years, several heuristic and data-driven algorithms have been developed for automatic EEG analysis by several research centers. In our brain-monitoring group, NeoGuard, tens of articles have been published and proposed different algorithms for automatic neonatal brain signal analysis, from very clinical points of view, to engineering methods, particularly based on machine learning algorithms. Recently, while a newly known set of algorithms, namely deep neural networks (DNN), was breaking new ground in various problems and areas of machine learning, we have extensively been working on its capabilities in EEG analysis and have developed three algorithms for seizure detection and sleep stage classification in preterm/term neonates. These algorithms can result in similar to better performance compared to the state-of-the-art algorithms, when they process the data much faster which is an important advantage for real-time monitoring.