From the results of our study, we identified significant positive correlations between CBF and cognition, when measuring using both ASL-MRI and SPECT. The main areas with significant results were located in PCC and temporo-parietal association cortexes, which are known to show decreases in perfusion or metabolism during cognitive decline in association with AD [
25‐
28]. These areas are the posterior parts of the default mode network (DMN), which has a primary network center in the PCC. These regions have a strong functional connection to the left and right inferior parietal lobule (IPL), ventral and dorsal medial prefrontal cortex, and lateral temporal lobes [
29]. Functional MRI studies have also consistently implicated the DMN as the most vulnerable network in AD [
30,
31]. The posterior (temporo-parietal-predominant) DMN may be particularly susceptible in early-stage AD [
32‐
34]. These studies therefore support our findings that a significant correlation exists between CBF and cognition, in the PCC and IPL. To the best of our knowledge, ours is the first report that succeeded in demonstrating such correlations between CBF and cognition using ASL-MRI, with supporting data using SPECT. These correlations may support the idea that regional CBF can serve as a biomarker of the neural changes underlying cognitive decline. As shown in Figs.
1b and
2b, the voxel values of SPM results dropped linearly with decreasing MMSE scores. Our results also showed similar significant correlations between cognition and CBF using ASL-MRI with PLD
1.5 and PLD
2.5, and SPECT. The measurement of CBF is thought to be influenced by modalities, tracers, and parameters. In fact, our results showed that ASL-MRI with PLD
1.5 and PLD
2.5 have significant differences. A significantly higher CBF was found at the adjacent areas of anterior cerebral arteries, middle cerebral arteries and posterior cerebral arteries for PLD
1.5 compared to PLD
2.5, suggesting early perfused areas. Liu et al. [
35] evaluated the CBF of AD patients using ASL-MRI with PLD
1.5 and PLD
2.5, and identified lower CBF for both PLD durations at the specific area of AD pathology when compared to healthy control subjects, but with smaller clusters of voxel for PLD
2.5. Despite of these significant differences in measured CBF using ASL-MRI with PLD
1.5 and PLD
2.5, PCC and temporo-parietal association cortexes were detected with significant correlations with cognition. This may suggest a possibility of the usefulness for the individual diagnosis using voxel-wise analyses of ASL-MRI. In Japan, voxel-wise analyses of SPECT using 3-Dimensional stereotactic surface projections (3D–SSP) [
36] and an SPM-based method termed “easy Z-score imaging system (eZIS)” [
37,
38] have been commonly used for the individual diagnoses in daily practices. Such voxel-wise methods may be helpful for making individual diagnoses using ASL-MRI. However, an age-specific normal database is required for the detection of significant abnormalities of individual images.
Our results also demonstrate a significant correlation between CBF and cognition at the right rectal gyrus for PLD1.5, and the right inferior temporal lobule and fusiform gyrus for PLD2.5. However, these areas are located at the edge of the brain, and these results might be caused by the errors during anatomical standardization and/or masking, and seem to be artifacts.
There were several limitations in this study. First, the sample size was limited; a larger number of subjects would be expected to provide more accurate results with correction for confounding variables, such as age and gender. Second, no images from cognitively normal subjects were used as controls, therefore we could not evaluate differences between AD patients and normal subjects. Third, the MNI template used for spatial normalization provided by SPM12 is based on a group of young, healthy individuals. The applicability of this template to older subjects is questionable. Fourth, the labeling efficiency plays a major role in PCASL quantification, but we did not calculate this parameter in this study. Furthermore, no conclusions could be drawn on the individual subject level, as the analyses were all performed on pooled data from each group.