The galaxy method for the 3-D spatial registration of PET images to Talairach brain atlas

Human brain function is studied by averaging many PET images in order to enhance the signal to noise ratio of weak group-specific patterns caused by medication, illness, or functional activation. Present registration methods by Woods and Friston (SPM), compute 3 translations and 3 rotations to register each PET image to a reference anatomical atlas using a 6 degree of freedom optimization method. By using only a 2 degree of freedom optimization method, the present method decreases the computing time down to 70 sec in a SUN Sparc2 CPU; this is a speed increment by a factor of 3 to 15 times over previous methods. Experimental results for 24 FDG-PET images also indicate that the present method improves the accuracy of the registration to a mean error of 0.48 mm and a maximum error of 2.47 mm.