Spatio Temporal Modelling plays an important role in personalized medicine, virtual clinical trials or drug target identification. It enables the encoding of trajectories of complex diseases, metabolic or developmental pathways, to optimise an individual’s disease treatment or determine a developmental status. Dynamic Developmental Patterns (DDP) form the main challenge in modelling trajectories, constituted of the incompleteness and irregularity of observations, inter-patient variability and impairing factors like comorbidity, age or individual treatment response. The focus of this thesis lies in providing new strategies for the spatio- temporal modelling of dynamic developmental patterns, to encode and understand baseline trajectories disentangled from time-dependent or systemic dynamics. Thus, on the one hand the identification of suitable baseline states is essential and on the other hand the development of techniques to analyse the dynamics’ deviations and relations to the baseline. Here, it is demonstrated that the proposed modelling concept is capable to flexibly model DDPs independent of the imaging modalities, of different populations/age ranges and applications to answer research questions in the field of computer vision, cancer research, brain development and functional connectivity network analysis. It leads to the development of novel data representation forms for DDPs, segmentation strategies, classification procedures and time-dependent prediction approaches, outperforming state of the art methods.
