A model-based approach for learning to identify cardiac arrhythmias

Abstract

ECG interpretation is used to monitor the behavior of the electrical conduction system of the heart in order to diagnose rhythm and conduction disorders. In this paper, we propose a model-based framework relying on a model of the cardiac electrical activity. Due to efficiency constraints, the on-line analysis of the ECG signals is performed by a chronicle recognition system which identifies pathological situations by matching a symbolic description of the signals with temporal patterns stored in a chronicle base. The model can simulate arrhythmias and the related sequences of time-stamped events are collected and then used by an inductive learning program to constitute a satisfying chronicle base. This work is in progress but first results show that the system is able to produce satisfying discriminating chronicles.