Each individual has a different size and shape of the brain. In order to compare data from several subjects, all the images need to be in the same space. In eZIS which employs SPM, this is accomplished by “normalizing” the images into the space defined by the Montreal Neurological Institute (MNI) template. The detail of normalizing process is described by Ashburner, one of the developers of SPM [6].

Following spatial normalization, the image data would be “smoothed” or blurred. By smoothing, the signal-to-noise ratio is increased. It also ensures that the residual differences are closer to Gaussian and decreases the intersubject registration error. Another reason for smoothing is simplifying the ensuing results, producing fewer regions of significant difference to report [7]. The outline of preprocessing is illustrated in Fig. 11.2.

After preprocessing, intensity of each image is normalized. Note that eZIS takes a different approach from SPM. In SPM, intensity normalization takes a two-step procedure. First, it calculates mean of an image. This averages all of the voxel values (VV) even outside the brain, so we need to threshold the image with certain value. Threshold is calculated by mean divided by 8. Then it calculates global value by averaging the VV above the threshold. Finally global value is set to 50 mL/min/100 g and each VV is adjusted with proportional scaling.

On the other hand, eZIS takes a straightforward procedure. First it applies a brain mask to preprocessed images. Then mean value is calculated with the masked images. Setting global value and adjusting VV is the same as SPM. Figure 11.3 illustrates how mean and VV of a certain voxel will change throughout the procedure. VV of a voxel coordinate [40 30 40] in the preprocessed image is 796.0 and mean VV is 368.8. Note that at this point, mean includes voxels outside the brain. After masking, VV at [40 30 40] remains the same, but the mean changes from 368.8 to 635.1. Now using proportional scaling eZIS sets mean as 50, VV at [40 30 40] will be Understanding this procedure is important because sometimes we see apparent hyperperfusion regions in eZIS results. For example, global cerebral blood flow (CBF) of advanced demented patients is generally low such as 30 mL/min/100 g, but the regional CBF in primary sensory motor region is intact such as 45 mL/min/100 g. In this case, after intensity normalization, the adjusted regional CBF in primary somatosensory regions will be which will result in remarkable hyperperfusion in this region. In order to prevent misinterpreting the results, it is recommended to check the global CBF regularly.

Following intensity normalization, z-score at each voxel is calculated with the following equation.

Here SD stands for standard deviation. This equation can be rephrased like the following.

So if Z-score is 2, one can say that subject’s VV in a certain voxel is 2 standard deviation below average of control subjects. This equation also tells us that we need to prepare the mean and standard deviation of control subjects beforehand. Z-score highly depends on the quality of control subjects, so we need to keep in mind that using different control subjects might result in quite a different result.

Z-score images are overlaid on the standard brain as shown in Fig. 11.1. Z-score threshold is an arbitrary set. In Fig. 11.1, z-score = 2 is set as a threshold.

SPECT findings of typical AD are characterized by hypoperfusion in (1) the parietotemporal association cortex, (2) posterior cingulate gyrus and precuneus; and (3) medial temporal areas [8]. It is also noteworthy that CBF of primary sensorimotor region is spared until the advanced stage. The yellow arrows in Fig. 11.4 point the characteristic hypoperfusion regions in AD and red arrow points the primary sensorimotor region where CBF is spared.