03:30 PM - 05:10 PM
Optimal Image Registration - From Applications to Code
Whenever images taken at different times, from different viewpoints, and/or by different sensors need to be compared, merged, or integrated, image registration is required. Image registration, also known as alignment, fusion, or warping, is the process of transforming data into a common reference frame and is essential for all imaging disciplines. Particularly in a medical environment, there is a huge demand for comparing pre- and post-intervention images, integrating modalities like anatomy (obtained, e.g., from computer tomography) and functionality (obtained, e.g., from positron emission tomography), motion correction and/or reconstruction of two-dimensional projections from a three-dimensional volume (applies to all tomography techniques and histology).
This tutorials motivates the registration problem and shows various fascinating medical applications. A flexible mathematical model based on a variational formulation is presented. Modelling aspects such as the design of application conform distance measures, regularization, and application dependent constraints are discussed. Following the path of the discretize then optimize paradigm, we discuss discretization which then leads to relative high dimensional numerical optimization problems. Exemplarily, we discuss generic registration (unconstrained), registration images showing soft and hard tissue (linearly constrained), and mass-preserving (non-linearly constrained) registration.
The tutorial is based on FAIR  (Flexible Algorithm for Image Registration) and the related freely downloadable MATLAB toolbox .
 J. Modersitzki, FAIR: Flexible Algorithms for Image Registration, SIAM, Philadelphia,