An Application of D ata Mining in Detection of M yocardial Ischemia utilizing pre- and post-Stress Echo Images PRAMOD K. SINGH Faculty of Information Technology University of Technology, Sydney PO Box 123,Broadway, NSW 2007, Australia Email: pksingh@it.uts.edu.au SIMEON J. SIMOFF Faculty of Information Technology University of Technology, Sydney PO Box 123, Broadway, NSW 2007, Australia Email: simeon@it.uts.edu.au DAVID D. FENG School of Information Techno logies, University of Sydney, NSW 2006, Australia Email: feng@it.usyd.edu.au Abstract: Automatic identification of endocardial and epicardial boundaries of LV has been a focus of research attention in the development of computational methods and computer support for cardiologists in identifying clinical heart disease and their diagnosis. Among heart imaging techniques, echocardiography offers significant advantages because of its low cost, portability, minimal discomfort, the absence of ionizing radiation, and its possible application for patient monitoring through real time processing. However, images generated from echocardiogram data are of poor quality. This paper presents the initial work in the development of a data mining approach for computer-assisted detection of myocardial ischemia, which includes Left Ventricle (LV) wall boundary identification, segmentation and further comparative analysis of wall segments in pre- and post stressechocardiograms. Keywords: Echocardiograms, Image processing, Multimedia Data mining, Object identification, Ischemia 1. Introduction The main objective of many efforts in cardiac imaging and image analysis is to access the regional function of the Left Ventricle (LV) of the heart. The general consensus is that the analysis of heart wall deformation provides quantitative estimates of the location and extent of Ischemic Myocardial Injury (1M!) [10]. Regional LV deformation can be determined using all of the principal imaging modalities, including contrast angiography, echocardiography, radio nuclide imaging, cine computed tomography (CT) and magnetic resonance (MR) imaging. Automatic identification of endocardial and epicardial boundaries of LV has been a focus of research attention in the development of computational methods and computer support for cardiologists in identifying clinical heart disease and their diagnosis. Echocardiography offers significant advantages over all other imaging techniques. The technique is attractive because of its low cost, portability, minimal discomfort, the absence of ionizing radiation, and its possible application for patient monitoring through real time processing [6, 11]. From a data mining point of view, data collected by echocardiograph systems includes sequence data of the heart behaviour. Myocardial ischemia is a heart disease induced by the obstruction of one or more coronary artery. LV is affected accordingly, which present the change of contractibility of certain segments of LV in echocardiograms images but very rarely on the whole ventricle. The abnormalities can be detected by detailed examination of the dynamics of each segment of LV walls and the coordination between them. Echocardiography is versatile; it may be combined with exercise, pharmacological, and other stressors and used in availability of circumstances less favorable to other techniques. The stress echocardiography provides a means of identifying myocardial ischemia by detection of stress-induced wall motion abnormalities by comparison of pre- and post stress images. The accuracy of stress echo cardiology in detecting significant coronary stenoses has proved to be from 80% to 90% depending on the population studies [11]. The technological revolution of ultrasound and digital technology brought this modality from a research to a clinical tool, but the interpretation of these studies remains still on subjective observation. From data mining point of view the echo data can be viewed as video data, which consists of a sequence of echo images, synchronized by the ECG signal. The basic requirement of quantitative analysis of echo images is the complete determination of inner (endocardial) and outer (epicardial) boundaries of the LV wall. In computer vision terms the finding of LV wall boundaries in echo images is an object detection problem. An object detection process typically involves image-processing algorithms for information extraction from images and further analysis of extracted information using priori knowledge of problem domain. A typical configuration of LV wall detection system is shown in Figure 1 [3]: MDMlKDD 2002: International Workshop on Multimedia Data Mining (with ACM SIGKDD 2002) 70