Evaluation of Five Algorithms for Mapping Brain Cortical Surfaces

With the increasing resolution and contrast of brain imaging devices automatic segmentation and quantification of the human cerebral cortex have grown popular for morphological analyses. The tightly folded cortex is often modeled with surfaces in 3D, and morphological features, such as the cortical thickness, can be calculated. In order to average and compare such morphological features within groups of subjects, mappings between the highly diverse cortical surfaces are needed. In this paper we evaluate five algorithms for mapping between discrete polygonal surfaces of cortices. Among the evaluated algorithms we include a new algorithm based on a functional expressing similarity between geometrical features. Four numerical mapping criteria, a landmark test, and statistical maps are used to evaluate the mapping algorithms. We show that the accuracy of manually placed landmarks are difficult to reproduce automatically, and the choice of mapping algorithm impacts the conclusions drawn from statistical maps generated by use of the algorithm. In terms of landmark accuracy, a spherical mapping approach with non-linear optimization is shown to be the best of the tested algorithms.