Using the Active Appearance Models for the Localisation of the Left Ventricle in Planar Scintigraphic Images.

Abstract - Statistical deformable models are powerful priori knowledge based technique for the delineation of the non rigid objects in images. The widely-known algorithms allowing such a priori information incorporation is the Active Shape Models (ASM) and the Active Appearance Models (AAM). In this paper, we focus on the detection of the left ventricle boundaries in planar nuclear imaging using AAM. In fact, this model cannot only incorporate the shape information but also the texture of the region of interest. Several tests were applied on a set of real images to optimize the choice of some model parameters. Results show the performance of the AAM method thus parameterized. Keywords: Cardiac nuclear images, left ventricle, deformable models, Active Appearance Models. 1 Introduction Nuclear medicine is a healthcare speciality involving the use of radioactive compounds to perform diagnostic imaging examinations that can lead to the effective treatment of many diseases. Although nuclear medicine is often considered as an independent discipline, it is closely related to radiology since radiation is used to develop images of human anatomy. Cardiac nuclear medicine refers to these diagnosis tests that are used to examine the heart function and anatomy. Several localization methods were applied to delimitate the left ventricle because of the importance of this muscle in pumping blood through the body. But, none of these techniques was sufficient to successfully detect the left ventricle boundary particularly with cardiac nuclear images. Other segmentation techniques using a priori knowledge about the ventricle are always proposed. This approach can boost the performance of the left ventricle delimitation. A widely-known algorithm allowing such a priori information incorporation is the statistical deformable models such as the Active Shape Models (ASM) method. This shape model was used by Khalifa and al. [1] to delimitate the left ventricle. However, in order to maximize the priori knowledge of the cardiac nuclear images, it is better to use grey-levels information, through objects of interest, further than only shape informations. Active Appearance Models (AAM) holds this information in consideration. In this paper, we try to delimitate the left ventricle boundaries using the AAM. In the second section, we present the advantage of using this method based on deformable models. In the third section, we describe the AAM method witch is interested in informations about both the shape and texture inside the objects. Then, we present its building steps and some examination tests.