Charité - Universitätsmedizin Berlin/DE
T. Chitiboi, J. Georgii, L. Goubergrits, M. Günther, A. Hennemuth, H. Meine, B. Menze, M. Neugebauer, B. Preim, T. Schäffter, M. Schwier, H. Mirzaee
Goals of the course
The course on MR image processing - from image data to information provides an overview on modern technologies for dealing with MR images. This ranges from simple pre-processing methods, over aligning dataset with different contrasts to quantitative analysis and visual exploration of results. A short outlook on using MR images for modelling is also given. The variety of methods that have been developed in the past is categorised and analysed critically. The course is aimed at providing the participants with criteria for deliberate selection of tools and methods in their studies. The course will provide practical tips and tricks for powerful processing of image data as well as several practical examples.
This course is intended for MR physicists, other scientists and PhD students who have already experience in basic MR methods and knowledge of MR acquisition principles, and who want to get a deeper insight into techniques for the assessment and comparison of image data, and the extraction and visual exploration of qualitative and quantitative image-based information.
The course will be held by an interdisciplinary team of experts in MR imaging, image processing, visualisation and mathematical modelling. It is organised in such a way that lectures are accompanied by hands-on sessions, so that participants do not only learn what techniques exist for the analysis and understanding of image information but also about the available tools and how the presented methods can be applied.
The program first introduces typical properties of MR image data that affect their analysis as well as methods for the quality assessment of image data. Techniques for the correction of typical errors (inhomogeneities, phase wraps, …) and the filtering of image data are presented and provided for the hands-on session.
The second part of the course introduces techniques for the extraction and quantitative assessment of anatomical structures such as vessels, organs or pathologies with special regard to reproducibility and robustness regarding intensity distribution, partial volume effects, anisotropic image resolution, etc. Available toolboxes such as Slicer, itk and MeVisLab will be applied in the accompanying hands-on session to test the presented approaches on real world data.
The training block on registration explains the basics of image registration for image fusion and motion analysis. Participants learn how to combine deformation/motion models with similarity measures and optimisation techniques to adapt classical registration approaches to their specific problems.
The introduction of these techniques is accompanied by a presentation of machine learning approaches for image post processing.
For the visual exploration of image data as well as derived information direct volume and surface visualisation techniques as well as methods from the field of visual analytics will be explained.
In an outlook session experts working in different clinical application fields will present examples on the coupling of image information with mathematical models.
• Which problems are caused by which imaging technique
• How to assess and enhance image quality: Linear and non-linear filtering
• Image enhancement methods
• Texture measurement (e.g. Vesselness filter)
Image segmentation and classification
• Interactive, semi-interactive and automatic segmentation methods à robustness vs. reproducibility
• Methods for the extraction of tubular structures
• Classifiers and their training (machine learning)