Skip to Main content Skip to Navigation

Prévision du risque neuro-développemental du nouveau-né prématuré par classification automatique du signal EEG

Abstract : The electroencephalogram (EEG), a measure of the electrical activity of the brain, remains one of the best non-invasive methods for predicting neurological outcomes. The aim of our work is to develop an automatic classification system which predicts risks on cerebral maturation that can lead to a pathological condition at 2 years. The EEG signal characteristics, which are useful for automated prediction, are processed via an application called EEGDiag, and applied to a set of 397 records for premature infants. Each record include an EEG record and a report on infant information and diagnosis at birth and 2 years later (normal, sick or risky). To assist physicians in preventing any abnormal neurological condition of the premature newborn, we have developed several intelligent classification models which can be applied to several series of characteristics of the EEG inspired from the annotations of neuropediatricians. Several classification and decisional aid models have been tested on different extracted databases in order to offer to doctors the best efficient classification system. Our proposed system automatically detects the prognosis of the premature newborn pathological condition. Our work consisted of subdividing the amplitude of EEG signal burst into three categories: low, medium and high. This subdivision study allowed to choose Intervals of these three categories which have served to greatly increase the performance of our intelligent classification system. A correlative data analysis allowed to create an independence and redundancy relation between the data attributes, which reduces the number of decisive parameters and thus selects the best series of parameters that made our system optimal and more efficient. These studies enabled us to achieve a classification system based on a series of 17 parameters with an accuracy 93.2%. This system can provide good sensitivity on predicting the neurological status of premature newborn and can be used as a decisional aid in clinical treatment.
Document type :
Complete list of metadatas

Cited literature [14 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Thursday, March 1, 2018 - 11:41:23 AM
Last modification on : Tuesday, March 31, 2020 - 3:20:14 PM
Long-term archiving on: : Wednesday, May 30, 2018 - 1:13:21 PM


Version validated by the jury (STAR)


  • HAL Id : tel-01720427, version 1


Yasser Alhajjar. Prévision du risque neuro-développemental du nouveau-né prématuré par classification automatique du signal EEG. Environnements Informatiques pour l'Apprentissage Humain. Université d'Angers, 2017. Français. ⟨NNT : 2017ANGE0020⟩. ⟨tel-01720427⟩



Record views


Files downloads