Automated VOI analysis in 18 F-FDDNP PET using structural warping: Validation through classification of Alzheimer's disease patients

In the field of quantitative imaging, the creation of accurate volumes of interest (VOIs) is often of central importance. However, the process of creating these VOIS for multiple subjects can be time-intensive and there are many chances to introduce variability on inter- and intra-investigator levels. Although pre­vious work has shown that image normalization through cortical surface mapping can be helpful in VOI analysis, the process is complicated and labor-intensive. In this paper we present a method to eliminate this variability by warping structural and functional images to a common space in which valid VOIs already exist. We apply this method to a study of Alzheimer's disease (AD) and 2-(1-{6-[(2-[18F]fluoroethyl)(methyl)amino]-2-naphthyl}ethylidene)malononitrile (FDDNP), which is known to co-localize with amyloid plaques and neurofibrillary tangles. We normalize the MRIs of control subjects and mild cognitive impairment (MCI) and AD patients, to a common space. The same normalization is applied to FDDNP PET images. The normalization technique reduces average voxel-to-voxel variance in MRIs by 54% as compared to linear normalization alone. Biologically important structures, such as the segmentation between white and gray matter, are maintained after normalization. Discriminant analysis shows that data extracted from VOIs in the common space out-performs data extracted from unnormalized PET images in classifying subjects as control, MCI, or AD. This suggests that image normalization may be useful in eliminating inter- and intra-investigator variability and increasing the predic­tive capability of data extracted from imaging modalities. Further study will examine the applicability of this method to predicting longitudinal changes in cognitive ability from functional imaging data.