Research On Personalized ECG Signal Automatic Classification Using Nonlinear Transformation

Cardiovascular disease,one of the top ranked killers in the world over the past decades,has been heavily investigated from different perspectives by the bioinformatics research community.Most of this high mortality rate is due to delayed diagnosis and late therapeutic interventions.Therefore,a continuous heart activity monitoring and timely prediction of potential heart abnormalities can improve significantly survival ratio,especially for high-risk patients.In this paper,a novel method is proposed to predict upcoming heart abnormalities by processing electrocardiogram(ECG)signals.The algorithm proposed in the paper is based on the cognitive process of human expert during decision making.The structure of the proposed system can be described by a cascade structure which consists of two layer of classifiers,namely,the Global Classifier and the Personal Classifier,which is responsible for personalized classification.The core idea behind the proposed method is to use a controlled nonlinear transformation to project extracted signal features into a higher-order dimensional space with desired geometric properties.In particular,the algorithm enforces the projected clusters of different abnormalities to symmetrically encircle the normal cluster through penalizing the clustering non-symmetry.This geometry is achieved by optimizing two objective functions.In this paper,the Multi-Objective Particle Swarm Optimization is applied for its optimum performance and converging speed over other algorithms.An immediate utility of this method is to characterize the deviation of ECG signal samples from the patient-specific norms towards different abnormality classes.Moreover,this method can be used to enhance our prediction about the potential upcoming heart problems before their occurrences.This is a critical point,since a timely diagnosis and therapeutic intervention can significantly reduce the heart related mortality rate.We applied this method to publicly available MIT-BIH dataset with three abnormality classes and the results suggest,respectively,8%,9%and 12%improvement in predicting the three abnormality classes.The proposed framework is general and applicable to a broad range of biomedical signals.