A comparison among several methods for building templates in functional brain imaging

Abstract

Spatial normalization of brain images to a common template is a usual preprocessing step in several comparison studies, brain tissue segmentation methods and computer aided diagnosis systems. In the [123I]FP-CIT SPECT and [18F]DMFP PET brain image modalities, the process of building the template is handicapped by the low-intensity values outside the striatum. In this paper, five MNI space templates are built for both image modalities without the coregistered MRI, a typical scenario in medical practice. Firstly, two standard methods are proposed based on the average of a set of SPECT and PET images. These images are previously coregistered to the MNI T1-weighted template and to a single functional image coregistered to the MNI template, respectively. Another approach iterates the registration and the average over the former templates. Finally, a posterization of the source images into three levels, striatum, non-specific (NS) regions and background, is performed following two procedures. In the fourth method, the posterized images are then used to weight the intensity values in the striatum, NS region and background to obtain a more accurate registration. This transform allows to register the source images to the MNI template with subsequent averaging. In the fifth method, the registration is performed between the posterized source images and the posterized MNI template, aiming to apply the obtained transformation to the original dataset. The posterized version of the MNI template is designed using the MNI atlas.