Simulation studies for the quantitative detection of lesions by statistical parametric mapping in assessing cerebral blood flow with SPECT

The purpose of this study was to determine whether SPM can effectively detect abnormalities on brain SPECT images. The averaged images of 9 normal control SPECT scans were generated using the MEAN function of an SPM´96. Artificial lesions were introduced on both sides of the temporal lobe by changing the mean local intensity on the average images for both a single slice and three slices. The intensity values were altered by -40%, -20%, +20% and +40% from normal values, for 20 or 40 voxels, and were examined in combination, i.e. both sides increased or decreased, or one side increased and the other decreased. The authors examined the effects of global normalization, and p- and extended p-values on quantified lesion sizes. SPM detected lesions for more voxels, higher intensity changes and multiple slices, better than lesions of less voxels, with lower intensity changes and a single slice. SPM detected increased lesions better than decreased lesions, when global normalization was applied. SPM localized lesions well, but global normalization and p- and extended p-values significantly effected the determined lesion sizes. The SPM analysis should be used with a full understanding of its potentials and limitations for clinical applications.