École Nationale Supérieure de
Techniques Avancées

Abstract

This report details my research during an internship at the Applied Mathematics Laboratory of École Polytechnique. The main objective of this internship was to perform Bayesian inference of an equivalent dipole model for the cardiac activity of a patient using surface recordings. I compared two approaches : the conventional Dowers method and Bayesian inference of Gaussian processes using the Kalman filter. The results demonstrated that the use of Gaussian processes provides better and more accurate predictions of cardiac signals. This advancement will have a significant impact by enabling more efficient detection of infarctions in patients. This research work contributes to the improvement of prediction methods and opens new perspectives in the field of cardiac condition detection based on surface recordings.