22. Comparison of the methods for brain parcellation

Abstract

In recent studies many scientists tend to examine functional connectivity with the help of graph theory algorithms, especially when studying resting-state fMRI data. The graph theory approach is based on the cortex parcellation – dividing it to disjoint areas and analyzing the relations between these areas. This work tries to compare results of parcellation by three types of anatomical atlases – AAL atlas, atlas of Brodmann areas and parcellation to gyri and sulci. We also use different approaches to compute representative signal for each area and compare how the representative explains data. We used mean signal, first, second and third principal component as the typical signal and we computed significances of each representative type by percentage of area variability explained by this signal. We found that the representative explains less than 50% of the variability in the area no matter which type of representative we chose. This could have been caused by present noise or the area size. The representative created from mean signal is almost the same as the first principal component signal regarding the percentage of explained variability. However, 1st PCA component can capture the inhomogeneity of the area and therefore can suggest possible dividing of the area which would result in smaller areas better explained by their separate representatives. Our comparison of anatomical atlases brought one compex conclusion – areas created according to Brodmann and AAL atlases have lower mean variance and higher mean explained variabillity than areas specified by atlas of gyri and sulci. To conclude we propose to use the AAL atlas above the other two types of atlases. In future the comparison of parcellation by anatomical atlases and data driven analysis would be interesting.