Validation of statistical parametric mapping (SPM) in assessing cerebral lesions: A simulation study.

Abstract

Simulated abnormalities were introduced in a normal SPECT with known and controllable characteristics (abnormality size and depth) in an attempt to provide validation for the analysis of SPECT lesion studies using SPM. Two simulations were carried out. The first determined the minimum hypoperfusion depth detectable using SPM by altering mean local intensity while keeping the size of the lesion constant. This was done by changing the mean local intensity in percentile increments of 10 down to -100 and up to 50. The second simulation determined the cluster size that SPM can detect by keeping the mean intensity of the lesion constant while altering its size from 4 voxels to 63,000 voxels in a total brain volume of 300, 000 voxels. Both simulations determined which method of normalization is most appropriate, what level of grey matter thresholding should be used, and at what statistical probability peak threshold (u) the results should be determined. Proportional scaling was found to be the most appropriate normalization method. ANCOVA was useful where very large abnormalities were present and normalization external to SPM was not available. In those cases, ANCOVA was used in conjunction with measurement of an unaffected part of the brain (in this case medial occipital lobe). For better results statistical probability peak threshold was set to p(u) = 0. 01 and grey matter threshold was set to a value below 0.5. SPM produced best results when the abnormality represented a decrease of about -50% from the normal or more and detected other decreases in an acceptable manner.