Bayesian Network with Decision Threshold for Heart Beat Classification

This work proposes a Dynamic Bayesian Network approach (BN) to support medical decision making in the problem of beat classification in electrocardiograms. The BN takes the uncertainty into consideration when making a decision every time new evidence is available. Moreover, the certainty related to the beat classification can be controlled through a threshold adjusted by the specialist. The performance of the BN based classifier is assessed through the MIT-BIH database, considering two beat classes: Premature Ventricular Beat (PVC class) and Other (gathering all other beat classes). The BN with probability threshold of 0.75 achieved scores of Sensitivity and Positive Predictive of 99% for the PVC beats. The results show that the BN framework is a promising tool for classifying cardiac arrhythmias.