INFARCTED LEFT VENTRICLE CLASSIFICATION FROM CROSS-SECTIONAL ECHOCARDIOGRAMS USING RELATIVE WAVELET ENERGY AND ENTROPY FEATURES

Abstract

Parasternal and apical echocardiography images captured from different cross-sectional planes (short-axis and four chambers) convey significant information about the structure and function of infarcted Left Ventricular (LV) myocardium. Thus, features from these cross-sectional views of echocardiograms extracted using computer-aided techniques may aid in characterizing Myocardial Infarction (MI). Therefore, this paper proposes a new algorithm for automated MI characterization using features extracted from parasternal short axis and apical four chambers cross-sectional views of 160 subjects (80 with MI and 80 normal) echocardiograms. The Stationary Wavelet Transform (SWT) method is used to extract the Relative Wavelet Energy and Entropy (RWE and RWEnt) features from the two cross-sectional views of echocardiography images separately. These features are ranked and subjected to classification in two different steps: (i) the features from each view are separately ranked using entropy, t-test and Wilcoxon ranking tests and fed to the classifier, and (ii) later, the features from both the views are combined and ranked. Finally, these ranked features are subjected to the Support Vector Machine (SVM) classifier for characterization of normal and MI using a minimum number of features. The proposed method is able to identify MI with 95.0% of accuracy, 93.7% of sensitivity and 96.2% of specificity using 32 features extracted from parasternal short-axis view; an accuracy of 96.2%, sensitivity of 97.5% and specificity of 95.0% with 18 apical four chamber view features. The results show that by combining the features from both views enables the confirmation of MI LVs with an accuracy of 96.8%, sensitivity of 93.7% and specificity of 100% using 16 features extracted from only two frames. Software development is in progress which can be incorporated into the echocardiography ultrasound machine for automated detection of MI patients.