Cardiac decision making using higher order spectra

Abstract The electrocardiogram (ECG) is the P-QRS-T wave representing the information about the condition of the heart. The shape and size of the ECG signal may contain useful information about the nature of disease afflicting the heart. However, these subtle details cannot be directly monitored by the human eye and may indicate a particular cardiac abnormality. Also, the ECG is highly subjective, the symptoms may appear at random in the time scale. Hence computer assisted methods can help physicians to monitor cardiac health easily and accurately. The ECG signal is nonlinear and non-stationary in nature. These subtle variations can be captured using non-linear dynamical Higher Order Statistics (HOS) techniques. Bispectrum is the third order spectra which captures information beyond mean and standard deviation. In this work we have analyzed five types of beats namely: Normal, Right Bundle Branch Block (RBBB), Left Bundle Branch Block (LBBB), Atrial Premature Contraction (APC) and Ventricular Premature Contraction (VPC). The extracted bispectrum features are subjected to principal component analysis (PCA) for dimensionality reduction. These principal components were fed to four layered feed forward neural network and Least Square-Support Vector Machine (LS-SVM) for automated pattern identification. In our work, we have obtained highest average accuracy of 93.48%, average sensitivity and specificity of 99.27% and 98.31% respectively using LS-SVM with Radial Basis Function (RBF) kernel. Our system is clinically ready to run on large amount of data sets.