Automated J Wave Detection Based On Hidden Markov Model

J waves are low-amplitude, high-frequency waveforms which look like notches or slurs appearing in the descending slope of the terminal portion of the QRS complex in electrocardiogram(ECG) of patients who are susceptible to cardiac arrhythmias and sudden cardiac death, after surviving early repolarization syndrome(ERS) or Brugada syndrome(BrS). Accordingly, J wave detection presents a non-invasive marker for some cardiac diseases clinically.In this dissertation, we first use the independent component analysis technology to extract the original J wave from ECG signal.And the extraction algorithm is improved by fuzzy neural network model. We define degree of separation.In the process of separation, according to the different degree of separation in the iterative process, improved algorithm choose different step size to balance the convergence speed and steady-state error in the extraction process.According to the extracted J wave, we construct ECG database of J wave detection of the next step. And then, we define five feature vectors including three time-domain feature vectors and two wavelet-based feature vectors,employing feature selection and applying principal component analysis(PCA)to reduce its dimensionality as an input of the classifier. Moreover, we discuss an optimal configuration of time-domain feature vector which is(a)140ms segment duration,(b)-20 ms segment location, and(c) retained nine PC. Finally,a Hidden markov model(HMM), trained by a proper set of these feature vectors,is employed as a classifier.The results show the proposed method, providing an average accuracy of93.8%, average sensitivity of 94.2%, specificity of 93.3% and average positive predict value of 93.4%, reveals high evaluation criteria(accuracy, sensitivity,specificity and positive predict value) and is qualified to detect J waves,suggesting possible utility of this approach for defining and detection of other complex ECG waveforms.