Skip to main content

An analyzer for preterm EEG

Periodic Reporting for period 1 - APE (An analyzer for preterm EEG)

Reporting period: 2015-09-01 to 2017-08-31

Compromised early brain development leads to lifelong disabilities, which have a heavy impact on the child, families, as well as society as a whole. Advances in clinical care have led to an increasing number of babies surviving extreme prematurity. The global challenge is to avoid early brain injuries by optimizing neurological care during the early days of neonatal intensive care. Improved neurological care during the early stages of life improves the long term health of the infant which reduces the social and economic burden to families and, therefore, society. This aim necessitates the use of constant, cot-side brain monitoring using the electroencephalograph (EEG), which faces formidable logistic challenges due to the need for large scale data analysis by EEG experts. The only imaginable solution for dealing with the vast amount of EEG information is to automate EEG analysis. This action proposes the development of an original, automated, cot-side Analyser for Preterm EEG (APE). We are focused on two particular tasks: 1) estimation of the conceptual age of the infant and 2) detection of abnormalities in the EEG or preterm infants. This algorithm will be based on the combination of state of the art biomedical signal processing techniques and recent advances in basic developmental neuroscience. An accurate cot side EEG analyser has strong clinical potential for improving early brain care, leading to lifelong improvements in affected individuals as well as unprecedented opportunities to benchmark new brain interventions.
Initial work conducted in the project involved the curation of two data sets of infant EEG recordings and associated demographic information. The first data set was recorded by collaborators at the Medical University of Vienna. It contained serial recordings of 60 infants from 26-38 weeks post-menstrual age (PMA) with multi-channel EEG. The second data set was recorded at the University of Helsinki and contained multi-channel EEG recordings from 80 infants. The University of Helsinki data set has been cleared for public release by the local ethics board and will be posted in late 2017/early 2018. These data sets were used to develop and train key algorithms for the estimation of brain maturation consistent with recorded EEG activity, detect neonatal EEG seizures (an abnormal activity) and identify artefacts in the EEG (contamination that is not associated with brain activity). The automated estimate of brain maturation could identify 80% of recordings with 2 weeks of PMA, a level of accurate that was preserved when reducing the number of recording electrodes from eight to two. The seizure detection algorithms detected 85-93% of seizures at a false detection rate of 1 per hour. The artefact detection system with a median area under the receiver operator characteristic of 0.950. These methods constitute a significant contribution to the development of methods for infant brain monitoring that generate more accurate and quantifiable measurements for assessing normal and abnormal behaviours. Result have been dissemination in international peer-reviewed journals and to the public via media releases from the University of Helsinki. These media release have been on-published in the local newspaper (Hufvudstadsbladet) and online publications such as med-gadget, e-health-news and the Daily Beast.
Methods that directly estimate the maturation of the brain in terms of age from EEG recordings did not exist before this project. The methods developed in this project represent progress beyond the state of the art in evaluation brain maturation in preterm infants. It is minimally invasive, highly applicable to the neonatal intensive care unit and directly targeted to measure brain function. It correlates highly with post-menstrual age of infants with normal neuro-developmental outcome. Methods of evaluating brain maturation are supported by additional algorithms developed in the project which show state-of-the-art performance for artefact and seizure detection. The next stage is to translate these algorithms into clinical practice as either single independent algorithms with a specific clinical use (such as seizure detection) and as a general multi-purpose neonatal EEG annotation system with a variety of uses. These methods provide quantitative measures of EEG activity which can be used as a short term outcome measure of neuro-development for clinical trials in the neonatal population. This measures complement neuro-psychological methods which have limited value in preterm infants.
The process of validation our EMA measurements