Visual Analysis of Heterogeneous Medical Data for Cohort Studies

Non-communicable diseases are responsible for the majority of deaths worldwide. Their burden on global health has been increasing steadily and will continue to do so in the following decade. In order to discover influencing factors for the development and treatment of such disease, we may study the distribution of the disease among populations. To this effect, in the recent years larger and more comprehensive population studies have been implemented. Numerous attributes ranging from selfreported interview data to results from various medical examinations are obtained from such studies resulting in large datasets where manual processing of the collected data is unfeasible. Approaches where experts do not need to process the data manually but also allow the interpretation of large amounts of data are thus desired. These methods should take into account the heterogeneous nature of the data. A holistic analysis of mixed data types, imaging data, and derived data such as extracted features should be supported by tools aiming to tackle risk or preventative factors to non-communicable diseases. In this thesis, methodology for the visual analysis of heterogeneous multidimensional data is presented. The proposed methods are able to support a wide range of tasks including but not limited to the identification of multidimensional trends, value retrieval, missing data analysis, and the generation of hypotheses for datasets. Quality metrics based on principal graph analysis are derived which allow the exploration of large numerical subspaces in order to provide medical insight. A range of applications were developed where features extracted from imaging data are used for aortic surgical planning, planar visualization of complex vascular structures, and analysis of ensembles of kidney segmentations. Real-world datasets are used to demonstrate the proposed approaches.