Positron emission tomography (PET) is a unique method to investigate
physiological processes in brain quantitatively. To be able to collect 
large databases of the functional data derived from PET new automatic 
methods for image analysis are needed. In this study we combine a new 
image reconstruction method, MRP Median Root Prior method, and new
image segmentation method, deformable model, to produce an automatic 
way to analyse PET image volumes. Effective noise reduction in image 
reconstruction is a prerequisite for good image segmentation with 
emission tomography images. MRP method has outstanding noise reduction 
properties without blurring of edges in the images. As an example of 
the segmentation the three-dimensional cortical structure was 
automatically extracted from 18 FDG PET images with the new deformable 
model. In the process no a priori information from anatomical images 
was applied. The average influx constants for FDG in the extracted 
structures of the 18 brain PET studies corresponded well to the
earlier findings with manual methods. The process produces the same 
surfaces and the same quantitative results in repeated segmentations 
provided that the initial settings for the segmentation are kept the 
same. In this study only the coarse three-dimensional volume of 
cortical structure was searched but the process can be done
iteratively so that in the next step the searching procedure could
 look for local features and substructures. Together with a good 
image reconstruction method the new image segmentation method gives 
a promising direction how PET brain images could be analysed 
automatically structure after structure.