Atrial fibrillation : detection of atrial rate and prediction during catheter ablation

This thesis treats of the problematic of patients with atrial fibrillation who undergo a catheter ablation. It comports two main objectives. The first one is to implement an algorithm capable of compute the atrial rate of a patient during the procedure from the detection of the A peaks on the patient's atrial rhythm signal. The second one is to find a way to predict the catheter ablation outcome (i.e. the absence of recurrence of the pathology) from the patient's characteristics. An empirical method has been implemented to detect the atrial rate, since the traditional frequency-based methods were unsuccessful. This method is based on finding the peaks of a signal and keeping the ones with some specificities. Those are determined by three parameters: the minimum height, the minimum peak separation and the minimum peak prominence. The first parameter is fixed, while the two other ones have to be selected. The user can either find these parameters manually (which gives a very good result) or use a suggested set of parameters (which can either give a good result or lead to a very big error). The learning has been conducted on 66 patients and 45 features. Among these features, the atrial rate and the maximum AA interval could not have been used, since those data where highly incomplete. Three algorithms have been considered: CART algorithm, Random Forest algorithm or $K$-Nearest Neighbours algorithm. The feature selection has been produced by two different methods: ReliefF and mutual information. Only the CART algorithm was able to output acceptable results with a BCR =0.73, while the other algorithms were not satisfying at all, giving BCR near 0.5 (total uncertainty)Master [120] : ingénieur civil en mathématiques appliquées, Université catholique de Louvain, 201