Ischemic beats detection using bispectrum analysis of the ischemic episodes in heart rate variability

In this paper, a new method for early and non-invasive diagnosis of myocardial ischemia is presented. Generally, clinical methods based on exercise tolerance test and angiography which is performed by feeding specific material in heart arteries is used for myocardial ischemia diagnosis. These methods are very invasive. Besides, morphological diagnosis of ST segment and T wave is very important and also difficult in Holter recording which is a long term electrocardiogram signal recording. In this survey automatic diagnosis of ischemia sections in long term ECG signal is done with bispectrum analysis of heart rate variability (HRV) signal which can be extracted from ECG signal and the bispectrum is obtained in episode time of 36 seconds. In this paper, 120 and 110 ischemia and normal episodes are investigated, respectively. Then, phase coupling in HF, LF and VLF bands are investigated. Average, variance and sum of bispectrum power in different bands of HRV signal are considered as input feature vector to the neural network. After that, classification is performed by KNN, PNN and SVM classifiers and by using leave one out cross validation method which has 85.09 % sensitivity, 91.74 % specificity and 88.35 % total accuracy.