On the 5th of November 2009 Julien Oster obtained the PhD degree es Sciences from the University of Nancy I.
Electrocardiogram (ECG) is required during Magnetic Resonance Imaging (MRI), for patient monitoring and for the synchronization of MRI acquisitions and heart activity. The MRI environment, due to its three characteristic physic components, highly disturbs ECG signals. For instance, the magnetic field gradients strongly complicate the ECG analysis in a non conventional manner. The development of specific signal processing tools is thus required. Existing methods, whether QRS detector or denoising techniques, do not accurately process these signals. A database of ECG acquired in MRI has been built, enabling the development of new processing techniques and their evaluation by using the folllowing two criteria : the cardiac beat detection quality and the signal to noise ratio estimated specifically on these particular recordings. A QRS detector, processing the noisy ECG signals, has been proposed. This technique is based on the singularity detection and characterisation provided by the wavelet modulus maximum lines. This detector provides helpful information on cardiac rhythm, for the development of novel techniques with a statistical approach. A new denoising method based on independent component analysis has been presented. This technique takes only advantage of the ECG signals. Two Bayesian based denoising methods, unifying two models (of ECG and gradient artifacts) in one state-space formulation have been proposed. Bayesian filtering has also been suggested for cardiac rhythm prediction, in order to improve the synchronization strategy.