ABSTRACT
The accurate prediction of neurodevelopmental outcomes of newborns suffering from hypoxic–ischemic cerebral injury during childbirth is vital to assist clinicians to correctly plan the newborn care and help enhance the newborn’s quality of life and reduce the social and financial costs. For newborns at high risk of prenatal brain injury, the multichannel electroencephalogram (EEG) is a valuable non-invasive screening, diagnostic, and monitoring tool. However, advances have been hampered by the amount of data and their complexity, the dependence on labor, and expertise-intensive manual processes involved in identifying and interpreting EEGs. A realistic way for dealing effectively with the big data formed by this vast amount of information is an intelligent knowledge-based computerized real-time system. The study reported in this chapter aims at minimizing the social and financial costs via early detection of brain disorders by extracting relevant patterns from the big data, therefore reducing the dimensionality of the problem and then automatically detecting newborn EEG features that best predict neurodevelopmental outcomes. This is achieved by developing high-resolution time–frequency (TF) techniques for EEG pre-processing, artefact detection/removal algorithms, and machine learning techniques. This study exploits the additional information provided by the nonstationarities of newborn EEG signals observed in the (t, f) plane and develops novel techniques for analyzing multichannel newborn EEG signals. Specifically, for the pre-processing of newborn EEG signals, this study proposes (1) a set of TF features for newborn EEG artefact detection and (2) a TF matched filtering technique and a TF blind source separation method for artefact removal. Finally, for the classification of newborn EEG abnormalities, classifiers based on support vector machines are designed that use TF features extracted from time–frequency representations of the multichannel EEG signals. The overall developed system can be implemented in a standalone EEG device or included as an add-on to existing medical devices. The research study described in this chapter provides a solution to the need for objective and automated detection and classification of newborn EEG abnormalities potentially resulting in a significant clinical impact.
