Stephan Witoszynskyj1, Alexander Rauscher2
1Department of Radiology , Medical
University of Vienna, Vienna, Austria; 2UBC MRI Research Centre,
University of British Columbia, Vancouver, BC, Canada
We
present a genetic algorithm for correction of motion artifacts in MRI. Two
types of genetic algorithms were investigated: the first used only
"non-sexual" multiplication and the second allowed
"cross-over" between solutions. The algorithm corrects for
translations by estimating correction factors for each k-space line. Four
different image metrics were studied: entropy, normalized-gradient-squared
(NGS), signal in the background and local coherence in the background. The
best results were obtained by using the simple algorithm and NGS and entropy
as metric. Since genetic algorithms are inherently parallelizable our
approach could benefit greatly from being implemented on computer clusters
and GPUs.