Fuzzy-based automatic detection of tags and ventricular contours for wall motion analysis of cardiac magnetic resonance images

We investigate the potential of a hybrid approach based on fuzzy set theory and image processing techniques to automatically detect tags and left ventricle contours in SPAMM cardiac images. Properties of the images such as brightness, contrast, and edges are extracted using conventional image processing techniques. Linguistic descriptions derived from "a priori" knowledge on tags and left ventricle contours are expressed through fuzzy membership functions, which fuzzify the feature space. The application of appropriate fuzzy relations and operators enables the classification of individual pixels of the images into four groups: tags, non-tags, myocardium and non-myocardium. Further feature and knowledge fuzzification allow the detection of the pixels belonging to tags, epicardium and endocardium. Snakes are employed to provide smooth and continuum curves.