Skip to Main content Skip to Navigation
Theses

Méthodes bio-inspirées appliquées à l’analyse temps réel des signaux biomédicaux

Mounia Hendel 1, 2
2 ABC - Machine Learning and Computational Biology
LORIA - ALGO - Department of Algorithms, Computation, Image and Geometry
Abstract : Brain computer Interfaces (BCI) give, through Electroencephalographic (EEG) signals, an opportunity for communication between the brain and the environment around a person with severe motor disability. Also, the electrocardiogram (ECG) is a graphic recording of the electrical activities of the heart. It provides important and useful information about the functional aspects of the heart. However, the implementation of BCI interfaces and the treatment of ECG signals need a good signal processing system, hows performance of depends mainly on the technique used to select the best characteristics, and the classification technique used to perform discrimination between different categories. We first apply a parameter extraction method based on the Discrete Wavelet Transform (DWT) and then we have extracted some statistical parameters. To avoid the overtraining and the curse-of-dimensionality in EEG signals application, we propose a new variable selection method based on Self Organizing Map (SOM) algorithm, which allows to group redundant and irrelevant parameters and select the better descriptors. Then, in the classification stage, we introduce the use of direct approach of probabilistic Multi-Class Support vector machines (M-SVM) in the problematic of EEG and ECG signals classification. Direct M-SVM are compared with two decomposition methods involving binary SVM based on approaches: one-against-one and one-against-all. Also, we post-processed the outputs of the M-SVM in terms of probabilities, to provide them with a calibrated reliability measure; and we investigated the possibility of implementing direct M-SVM in real-time systems. The proposed approach is validated on the EEG and ECG signals taken from the databases: "Keirn and Aunon" and "Mit-Bih" respectively. The results demonstrated that direct M-SVM achieve better average precision than indirect M-SVM. In addition, the results demonstrated that direct M-SVMs can be embedded in real-time systems.
Keywords : ECG EEG M-SVM SOM DWT BCI
Document type :
Theses
Complete list of metadatas

https://hal.univ-lorraine.fr/tel-03042290
Contributor : Mounia Hendel <>
Submitted on : Sunday, December 6, 2020 - 1:34:26 PM
Last modification on : Wednesday, December 9, 2020 - 3:39:17 AM

File

thése-final-Hendel.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : tel-03042290, version 1

Citation

Mounia Hendel. Méthodes bio-inspirées appliquées à l’analyse temps réel des signaux biomédicaux. Informatique [cs]. Université Mohamed Boudiaf des Sciences et de la Technologie - Mohamed Boudiaf d'Oran, 2017. Français. ⟨tel-03042290⟩

Share

Metrics

Record views

42

Files downloads

26