Abstract:

MR imaging is based on encoding spatial information onto the observed spins by use of gradients. The signal is measured as a superposition of spatial frequencies in a domain called k-space, which is related to the image domain by Fourier Transformation. For conventional imaging based on rectilinear sampling of k-space data basic image properties are defined by the Nyquist-Shannon theorem, which sets stringent limits with respect to the relationship between spatial resolution, sampling time, and volume coverage irrespective of the exact way of signal generation. Parallel imaging techniques and inverse reconstruction based on highly undersampled data allows to alleviate these limits at least to some extend and thus to accelerate data acquisition tremendously.

Images are not only pictures, images represent spatially distributed information. An increasingly important task is to find ways to extract the information content from the images and to make it available for further use in research and diagnosis. The complexity of tasks can vary from simple routines to improve image quality to the extraction of physiological and functional parameters from multiple images to multimodal image analysis including non-imaging information using machine learning approaches. The presentation will show examples of basic and advanced data processing and also present concepts for integration of advanced processing tools into the data handling workflow for easy application by non-expert users.