Skip to content


Jan Hering
Robust motion and artifact correction in diffusion-weighted MR images
On 2016-05-24 11:00 at G205
Diffusion-weighted Magnetic Resonance Imaging (MRI) is an image acquisition
technique, which allows for the mapping of molecular diffusion in biological
tissues, reflecting the micro-structural tissue properties and thus offering a
unique non-invasive insight into the human brain.

A diffusion MRI acquisition consists of many three-dimensional volumes, each of
them measuring the tissue diffusivity in a different pre-defined direction
(gradient) and strength (b-value). The high amount of acquired images leads
inevitably to longer acquisition times and increased subject's motion. In
addition, the acquisition is susceptible to artifacts caused by eddy-currents
emerging during scanning. Thus, effective image-based artifact correction is an
essential step in the analysis of diffusion MR images since the expected
sensitivity and specificity with regard to micro-structural effects can only be
achieved if the different volumes are exactly aligned to each other.

Many current methods are based on pair-wise image registration, which becomes
challenging with increasing contrast differences and low signal-to-noise ratio
at high b-values. In contrast to these methods, which use a single image
the presented method allows for simultaneous exploitation of the different
signal intensity relationships between the image volumes. To achieve so, the
presented multi-objective method follows a memetic search scheme, a hybrid
meta-heuristic optimization principle.  The achieved robustness and precision
the approach as well as the impact on further diffusion MRI analysis steps like
fiber-tracking could be demonstrated on both synthetic and in-vivo datasets and
will be also addressed during the talk.

One characteristic feature of meta-heuristic optimization approaches is their
transferability to other classes of problems. Hence, the application of the
presented multi-objective registration method is not restricted to diffusion
pre-processing, but could be applied to other image-registration problems with
multiple intensity relationships between the images.
Back to the list