Skip to Main content Skip to Navigation
Theses

Classification des signaux EMG utérins afin de détecter les accouchements prématurés

Abstract : Premature birth remains the main cause of neonatal mortality and morbidity. Uterine EMG signal seems a potential vector for the indication of the risk of premature birth.
In the continuation of the work completed for detection, the treatment and the classification of the events in Uterine EMG signal, our work was dedicated to the classification of the EMG contractions, in order to part the two types of labor: premature deliveries and deliveries at term.
The uterine contractions were manually segmented from the uterine EMG signals. Then each contraction is modeled and the parameters are extracted before making the classification. This modeling is made by wavelet and the analysis of the power spectral density of each contraction. Classification is then carried out by using 2 types of methods: first a non-supervised classification method, which is used to group the contractions with no a priori knowledge of the classes, and then permits to make an interpretation of the groups according to the weeks of gestation and term of delivery. In this context we have developed an original method of non supervised classification based on the Fisher test combined with the K-mean method (USCM, Unsupervised Statistical Classification Method).
The other type of used classification is supervised. After having selected in a precise way the women who can be used for the training of our method of classification, we used various supervised methods of classification. First, we used traditional methods (neural Networks, Parzen...). Then an original method based on the Wavelet network has been developed for this classification. This method had been previously used for the regression but never for classification.
We were faced with the low number of the set of training. We thus also developped a method based on autoregressive model (AR Model) to increase the training set.
Concerning the applications, and for separation between the EMG signals (clinical application), we used two approaches. In the first approach, we used contractions having the same RWG (Registration Week of Gestation) but different BWG (Birth Week of Gestation), by testing small and large differences. The second approach is to classify the events acquired with different RWG for women having the same BWG.
According to the results obtained, we can conclude that we can distinguish different delivery terms from recordings obtained from women having the same term of pregnancy. We can also conclude that the contractions change their characteristics according to the term of pregnancy.
In a clinical point of view, the important result is that we could distinguish, for a given recording term, normal contraction to contractions leading to the premature delivery.
Document type :
Theses
Complete list of metadatas

Cited literature [147 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00410409
Contributor : Chantal Guilbert <>
Submitted on : Thursday, August 20, 2009 - 2:11:46 PM
Last modification on : Thursday, January 11, 2018 - 6:17:20 AM
Long-term archiving on: : Monday, October 15, 2012 - 4:15:49 PM

Identifiers

  • HAL Id : tel-00410409, version 1

Collections

Citation

M.O. Diab. Classification des signaux EMG utérins afin de détecter les accouchements prématurés. Sciences de l'ingénieur [physics]. Université de Technologie de Compiègne, 2007. Français. ⟨tel-00410409⟩

Share

Metrics

Record views

467

Files downloads

3675