Identification of myocardial infarction tissue based on texture analysis from ultrasound images

Abstract

Texture is an important characteristic that can be used for identification and/or detection for surface defects or abnormalities. This research has developed an algorithm for identifying heart with suspected myocardial infarction problem based on texture analysis applied on echocardiography images. A hybrid technique of wavelet extension transform with gray level co-occurrence matrix is proposed. In this work wavelet extension transform is used to form an image approximation with higher resolution. The gray level co-occurrence matrices computed for each subband are used to extract four feature vectors: entropy, contrast, energy (angular second moment) and homogeneity (inverse difference moment). The classifier used in this work is the Mahalanobis distance classifier. The method is tested with clinical data from echocardiography images of 30 patients. For each patient, tissue samples are taken from suspected infarcted area as well as from non infarcted (normal) area. For each patient, 10 image frames separated by some time interval are used and for each image frame 5 normal regions and 5 suspected myocardial infarction regions of 16x16 pixel size are analyzed. The proposed method has achieved 91.67% performance accuracy in classifying between normal and infarcted hearts. Thus, the proposed technique may be used as a computerized second opinion for determining whether a person is suffering from a myocardial infarction heart or not.