Classification de séries temporelles par modèles markoviens cachés, application à l'ischémie myocardique

Abstract

A new method for myocardial ischemia detection is proposed in this communication. The originality of this method relies on the consideration of the dynamics of times series extracted from the ECG whereas traditionnal methods are simply based on static measures. After the extraction of a feature vector, the dynamics are caracterised with an Hidden Semi-Markovian Model (HSMM). The ischemic detector uses a reference HSMM and an ischemic HSMM and then compare the likelihood of the time series. Results obtained with PTCA (percutaneous transluminal coronary angioplasty) records of the STAFF3 database show a very good detection rate (96% of sensibility and 80% of specificity).