Genetic algorithms applied to Fourier-descriptor-based geometric models for anatomical object recognition in medical images

摘要

Abstract: In this work we encode the shape complexity of a searchobject using 3D Fourier descriptors (FDs) and allowgenetic algorithms (GAs) to optimize the object's shapeand position. Using magnetic resonance image (MRI)data, we perform an approximate segmentation on onelateral ventricle in the brain and use the FDs fromthis as seeding values for the GAs to search for theleft and right lateral ventricles in subsequent 3D datasets. We show that the method is capable of coping withnormal biological variation. We compare a GA-guidedsegmentation with an interactive region growing methodand find an agreement of not less than 80 plus or minus6% in voxel classification with a corresponding averageedge placement error of 2.2 plus or minus 0.4 mm.Finally we examine how the optimization can be speededup by a distributed parallel implementation. !31