"Recent hardware developments from the medical device manufacturers have made possible non-invasive/in-vivo acquisition of anatomical and physiological measurements. One can cite numerous emerging modalities (e.g. PET, fMRI, DTI). The nature (3D/multi-phase/vectorial) and the volume of this data make impossible in practice their interpretation from humans. On the other hand, these modalities can be used for early screening, therapeutic strategies evaluation as well as evaluating bio-markers for drugs development. Despite enormous progress made on the field of biomedical image analysis still a huge gap exists between clinical research and clinical use. The aim of this proposal is three-fold. First we would like to introduce a novel biomedical image perception framework for clinical use towards disease screening and drug evaluation. Such a framework is expected to be modular (can be used in various clinical settings), computationally efficient (would not require specialized hardware), and can provide a quantitative and qualitative anatomo-pathological indices. Second, leverage progress made on the field of machine learning along with novel, efficient, compact representation of clinical bio-markers toward computer aided diagnosis. Last, using these emerging multi-dimensional signals, we would like to perform longitudinal modelling and understanding the effects of aging to a number of organs and diseases that do not present pre-disease indicators such as brain neurological diseases, muscular diseases, certain forms of cancer, etc.Such a challenging and pioneering effort lies on the interface of medicine (clinical context), biomedical imaging (choice of signals/modalities), machine learning (manifold representations of heterogeneous multivariate variables), discrete optimization (computationally efficient inference of higher-order models), and bio-medical image inference (measurement extraction and multi-modal fusion of heterogeneous information sources)."