Segmentation of Medical Images Applications in Echocardiography and Nuclear Medicine

Segmentation is an important task in all kinds of image analysis. In medical image analysis segmentation has a great clinical value since the aim could be to localize organs or pathologies in order to raise the quality of diagnoses. This thesis consists of three papers where different segmentation techniques have been applied to different imaging modalities. In the first paper a full computer assisted diagnosis system is presented where the aim is to find abnormal lesions in scintigraphy images of the kidneys. Here we have two segmentation parts; first segmentation of each kidney in the images using an active shape model and then localization of potential lesions using thresholding. In a test group of 56 patients the segmentations work very well just like the classification of lesions does. The second application is segmentation of the left ventricle in echocardiographic images. This segmentation is important when measuring the left ventricular function. The segmentation is done using a region based snake where the data term is driven by virtual image forces derived from the image intensities. To overcome problems with the cardiac valve opening and closing during the cardiac cycle, we annotate two anchor points, one on each side of the valve. We track these through the cycle in order to minimize user interaction and no segmentation is done over the valve. This method shows promising results. In the third paper we have developed a measure of the shape of the septum, the wall between the left and the right ventricle, to be used in echocardiographic images of patients with a mechanical pump attached to their heart as a bridge to transplantation. Such a measure can be useful when tuning the speed of the pump. Here the segmentation of the septum is achieved using the shortest path algorithm. The septum measure is then a measure of how much it deviates from a straight septum. Our septum measure corresponds in most cases to the assessments from a physician. These three applications show the usefulness of segmentation in a variety of applications within medical imaging.