Accuracy and Precision of the Computerized Brain Atlas Programme for Localization and Quantification in Positron Emission Tomography

Abstract

The computerized brain atlas programme (CBA) provides a powerful tool for the anatomical analysis of functional images obtained with positron emission tomography (PET). With a repertoire of simple transformations, the data base of the CBA is first adapted to the anatomy of the subject's brain represented as a set of magnetic resonance (MR) or computed tomography (CT) images. After this, it is possible to spatially standardize (reformat) any set of tomographic images related to the subject, PET images, as well as CT and MR images, by applying the inverse atlas transformations. From these reformatted images, statistical images, such as average images and associated error images corresponding to different groups of subjects, may be produced. In all these images, anatomical structures can be localized using the atlas data base and the functional values can be evaluated quantitatively. The purpose of this study was to determine the spatial and quantitative accuracy and precision of the calculated regional mean values. Therefore, the CBA was applied to regional CBF (rCBF) measurements with [¹¹C]fluoromethane and PET on 26 healthy male volunteers during rest and during three different physiological stimulation tasks. First, the spatial accuracy and precision of the reformation process were determined by measuring the spread of defined anatomical structures in the reformatted MR images of the subjects. Second, the mean global CBF and the mean rCBF in the average PET images were compared with the global CBF and rCBF in the original PET images. Our results demonstrate that the reformation process accurately transformed the individual brains of the subjects into the standard brain anatomy of the CBA. The precision of the reformation process had an SD of ∼1 mm for the lateral dislocation of midline structures and ∼2–3 mm for the dislocation of the inner and outer brain surfaces. The quantitative rCBF values of the original PET images were accurately represented in the reformatted PET images. Moreover, this study shows that the application of the CBA improves the analysis of functional PET images: (a) The average PET images had a low background noise [0.4 ml/100 g/min ± 0.7 (SD)] compared to the mean rCBF changes specifically induced by physiological stimulation. (b) The reformatted PET images had a voxel volume of 10.9 mm³. Owing to this high sampling resolution, it was possible to differentiate the mean rCBF changes in adjacent activated fields such as the left motor hand area from the sensory hand area and the left premotor cortex. (c) By calculating the relation of the mean rCBF change to the SEM of the mean rCBF change on a pixel-by-pixel basis, areas with significant rCBF changes could be determined. By use of the CBA, it was found that there was a high intersubject consistency in location of stimulation-induced rCBF changes. Furthermore, the rCBF changes in specifically stimulated areas were of similar magnitude among the subjects. It was shown that the stimulation-induced mean rCBF increases may be accompanied by mean rCBF decreases in other areas.