Novel Approaches to the Representation and Analysis of 3D Segmented Anatomical Districts

Nowadays, image processing and 3D shape analysis are an integral part of clinical practice and have the potentiality to support clinicians with advanced analysis and visualization techniques. Both approaches provide visual and quantitative information to medical practitioners, even if from different points of view. Indeed, shape analysis is aimed at studying the morphology of anatomical structures, while image processing is focused more on the tissue or functional information provided by the pixels/voxels intensities levels. Despite the progress obtained by research in both fields, a junction between these two complementary worlds is missing. When working with 3D models analyzing shape features, the information of the volume surrounding the structure is lost, since a segmentation process is needed to obtain the 3D shape model; however, the 3D nature of the anatomical structure is represented explicitly. With volume images, instead, the tissue information related to the imaged volume is the core of the analysis, while the shape and morphology of the structure are just implicitly represented, thus not clear enough. The aim of this Thesis work is the integration of these two approaches in order to increase the amount of information available for physicians, allowing a more accurate analysis of each patient. An augmented visualization tool able to provide information on both the anatomical structure shape and the surrounding volume through a hybrid representation, could reduce the gap between the two approaches and provide a more complete anatomical rendering of the subject. To this end, given a segmented anatomical district, we propose a novel mapping of volumetric data onto the segmented surface. The grey-levels of the image voxels are mapped through a volume-surface correspondence map, which defines a grey-level texture on the segmented surface. The resulting texture mapping is coherent to the local morphology of the segmented anatomical structure and provides an enhanced visual representation of the anatomical district. The integration of volume-based and surface-based information in a unique 3D representation also supports the identification and characterization of morphological landmarks and pathology evaluations. The main research contributions of the Ph.D. activities and Thesis are: • the development of a novel integration algorithm that combines surface-based (segmented 3D anatomical structure meshes) and volume-based (MRI volumes) information. The integration supports different criteria for the grey-levels mapping onto the segmented surface; • the development of methodological approaches for using the grey-levels mapping together with morphological analysis. The final goal is to solve problems in real clinical tasks, such as the identification of (patient-specific) ligament insertion sites on bones from segmented MR images, the characterization of the local morphology of bones/tissues, the early diagnosis, classification, and monitoring of muscle-skeletal pathologies; • the analysis of segmentation procedures, with a focus on the tissue classification process, in order to reduce operator dependency and to overcome the absence of a real gold standard for the evaluation of automatic segmentations; • the evaluation and comparison of (unsupervised) segmentation methods, finalized to define a novel segmentation method for low-field MR images, and for the local correction/improvement of a given segmentation. The proposed method is simple but effectively integrates information derived from medical image analysis and 3D shape analysis. Moreover, the algorithm is general enough to be applied to different anatomical districts independently of the segmentation method, imaging techniques (such as CT), or image resolution. The volume information can be integrated easily in different shape analysis applications, taking into consideration not only the morphology of the input shape but also the real context in which it is inserted, to solve clinical tasks. The results obtained by this combined analysis have been evaluated through statistical analysis.