Segmentation of the right ventricle in cardiac CT images

Abstract: Using cardiac computed tomography (CT), images of both the left and right ventricle can be acquired. Potential findings of interest are for example enlarged ventricles or inadequate pumping, which can be used for diagnosis. The analysis requires that the ventricles are outlined, either manually or automatically. Automated segmentation methods for clinical use exist for the left ventricle but not for the right ventricle. Automated methods are necessary in clinical routine as manual outlining in hundreds of images per patient is time consuming. Automated methods would also allow to determine reference values for the right ventricle volume in CT images which currently is missing. Therefore, the purpose of the thesis was to develop and validate an automated method to segment the right ventricle. In this thesis an active shape model is implemented and validated for automatic segmentation of the right ventricle in cardiac CT images. The model was trained on sixty subjects who all underwent a magnetic resonance (MR) examination and had their left and right ventricles manually outlined by professional physicians. Validation was performed on patients with varying image quality from the Heart and Lung-foundation founded SCAPIS research project. The developed method has proven flexible enough to handle the many shapes of the right ventricle while not being tied to good image quality or strong edges. The algorithm will be implemented in the first version of the clinical software Segment CT provided by Medviso AB.