Effects of Age on the Glucose Metabolic Changes in Mild Cognitive Impairment

PET and MR Imaging Acquisitions

All subjects underwent FDG-PET and MR imaging studies within 110 days of the clinical evaluation. FDG-PET scans were performed on the Advance PET scanner (GE Healthcare, Milwaukee, Wisconsin) operating in 3D mode. The subjects were injected with FDG in a dimly lit room with minimal auditory stimulation. After a 30-minute uptake period, PET acquisition was followed by a transmission scan. Emission data were reconstructed into a 128 × 128 matrix with a 256-mm FOV. The pixel size was 2.00 mm with a 4.25-mm section thickness. All subjects underwent MR imaging on a 1.5T scanner (Signa, GE Healthcare). A T1-weighted 3D volumetric spoiled gradient-recalled echo sequence with 124 continuous partitions in coronal plane, a 1.6-mm section thickness, a 24 × 18.5 cm FOV, 192 views, and a 25° flip angle was acquired for registration and segmentation of FDG-PET images. A FLAIR pulse sequence in the axial plane (TR/TI/TE = 16,000/2600/140 ms, 256 × 160 matrix, 1 repetition, 22-cm FOV, 3-mm interleaved images) of the whole head was used for the assessment for cerebrovascular lesions.

Assessment of Cerebrovascular Lesions

Cerebrovascular lesions were identified and rated on FLAIR images because they contribute to cognitive impairment in MCI.¹³ A radiologist (K.K.), blinded to all clinical information, assessed the WMH and hemispheric cortical and lacunar infarctions. WMH volume was estimated by visually comparing the subject's FLAIR images with a bank of 10 FLAIR image templates with increasing WMH volumes (from 1 to 100 cm³) determined with an automated image-segmentation algorithm.¹³ The WMH volume-estimation algorithm was previously validated against quantification by using automated image segmentation of the WMH volume.¹⁴ Because cortical infarctions of >1 cm in the largest diameter were one of the exclusion criteria, we only evaluated hemispheric cortical infarctions ≤1 cm in largest diameter. Subcortical infarctions were defined as discrete subcortical lesions of >3 mm in diameter with an intensity equivalent to CSF on FLAIR images and an accompanying hyperintense gliotic rim.¹³

FDG-PET Analysis

The AAL atlas¹⁵ was modified in-house to contain the following labeled regions of interest: right and left parietal lobes that included the posterior cingulate gyrus and precuneus, temporal lobe, and pons. These regions were chosen on the basis of previous reports showing decreased FDG uptake in these regions in patients with pathologically proved AD,^16–18 and the pons was chosen as an internal reference region of interest.⁶ The high-resolution T1-weighted single-subject brain image¹⁵ with atlas labels was normalized to the custom template generated from 200 subjects with AD and 200 CN subjects¹⁹ by using the unified segmentation method in SPM5 (Wellcome Department of Imaging Neuroscience, London, United Kingdom),²⁰ giving a set of labels corresponding to the custom template space. The resulting region-of-interest labels were then imported into the Analyze (Mayo Clinic, Rochester, Minnesota) region-of-interest tool and hand-edited by a trained technician (M.M.S.), reviewed by a radiologist (K.K.), and saved to produce a more accurate set of region-of-interest labels corresponding to the custom template, which we refer to as the “custom template AAL” atlas. The FDG-PET image volume of each subject was coregistered to his or her own T1-weighted MR imaging scan, by using a 6-degrees of freedom affine registration with a mutual-information cost function.

Each subject's MR image was then spatially normalized to the custom template by using the unified segmentation model of SPM5,²⁰ giving a DCT, say G_i, which normalizes the MR image of subject i to the custom template. Then for each subject, the inverse transformation (G_i⁻¹) was applied to the custom template AAL atlas to warp the atlas labels to the subject's native anatomic space. Atlas-based parcellation of FDG-PET images into regions of interest was, therefore, performed in subject space. For each subject, the native-space segmented gray and white matter probability maps generated from the unified segmentation routine were combined to create a binary brain mask. Each subject's brain mask was then multiplied by the subject-specific warped atlas, to generate a custom atlas for each subject, parcellated into the aforementioned regions of interest. This step was performed to minimize inclusion of nonbrain regions in statistics of all regions of interest.^21,22 Partial volume correction for tissue and CSF compartments was applied by using the 2-compartment model.²¹

Statistics on image voxel values were extracted from each labeled cortical region of interest in the atlas. FDG ratio images were calculated by dividing the median value in each target cortical region-of interest value by the median value in the pontine region of interest of the atlas.⁶ A temporoparietal cortical FDG uptake summary measure (temporoparietal FDG) was formed by combining the left and right parietal lobes (including the posterior cingulate gyrus and precuneus) and temporal lobe ratio values for each subject. The temporoparietal FDG was computed as the ratio of the median FDG uptake of the contributing regions of interest to the median FDG uptake of the pontine region of interest. Temporal and parietal regions were chosen to derive the temporoparietal FDG measure because decreased glucose metabolism in these regions is a sensitive marker for AD-related pathology.²²

Statistical SPM5 was used to evaluate FDG uptake on a voxelwise basis.²⁰ For each subject, all voxels in the subject's FDG-PET image were divided by the median FDG uptake of the pontine region of interest to form uptake ratio images. Then for each subject, i, spatial normalization of the FDG-PET uptake ratio image to custom template space was performed by using the DCT normalization parameters, Gi, obtained as described above. Finally, smoothed with a Gaussian kernel with a full width at half maximum of 8 mm using these smoothed images, voxelwise FDG uptake differences between groups were assessed in SPM5. Statistical maps of group differences were displayed at a significance value of P < .005, uncorrected for multiple comparisons.

VBM

First, a custom template and tissue probability maps were created in SPM5, by using the T1-weighted MR images of the 25 subjects with aMCI and 25 CN subjects.²³ The custom template and tissue probability maps were created by first normalizing and segmenting the 50 scans by using the unified segmentation model in SPM5 with the standard Montreal Neurologic Institute template and tissue probability maps, followed by a cleanup step that uses a hidden Markov random field model to increase the accuracy of the individual subject's tissue probability maps, and finally by averaging the normalized subject tissue probability maps. All subject images were then normalized and segmented by using the unified segmentation model and the custom tissue probability maps, followed by the hidden Markov random field cleanup step. The Jacobian modulation was applied to compensate for the effect of spatial normalization and to restore the original absolute gray matter volume in the segmented gray matter images. These modulated images were then smoothed with a 6-mm full width at half maximum smoothing kernel. Gray matter differences between groups were assessed by using a 2-sided t test within the general linear model framework of SPM. Group differences were displayed at a significance value of P < .005, uncorrected for multiple comparisons. Note that the same significance threshold was used for the FDG-PET analysis.

Statistical Analysis

Wilcoxon rank sum tests were used for statistical analysis comparing the demographic and clinical features of the clinical groups within age strata. Various types of cerebrovascular lesion loads were compared with Kruskal-Wallis or the Fisher exact tests in the 4 groups. Age effects on WMH were tested by using the Spearman rank-order correlation to test for a relationship between age, as a continuous measure, and FDG-PET uptake within clinical groups. Additionally, a linear regression model was fit with main effects for age and clinical diagnosis group as well as an interaction term between age and clinical group. This interaction term was used to assess different associations with FDG-PET and age among the clinical groups.