The Association of Metabolic Brain MRI, Amyloid PET, and Clinical Factors: A Study of Alzheimer's Disease and Normal Controls From the Open Access Series of Imaging Studies Dataset

Although brain activities in Alzheimer's disease (AD) might be evaluated MRI and PET, the relationships between brain temperature (BT), the index of diffusivity along the perivascular space (ALPS index), and amyloid deposition in the cerebral cortex are still unclear.

I maging techniques plays an important role in the diagnosis of Alzheimer's disease (AD).As MRI and positron emission tomography (PET) technologies have evolved, brain metabolism has been potentially assessed in various ways.4][5][6] Amyloid PET, including 18 F-florbetapir, 18 F-flutemetamol, and 18 F-florbetaben as nuclides, allows in vivo detection of amyloid plaques, one of the main neuropathological landmarks of AD, with very high sensitivity and specificity. 7,8The sensitivity of amyloid PET for predicting progression from mild cognitive impairment to AD has been reported to be higher than that of 18 F-FDG PET, and it has gained worldwide adoption. 9he relationship among these imaging metrics, age, and cognitive status remains controversial.Sparacia et al. 10 found no significant BT differences between patients with AD and normal subjects, while Asfuro glu et al. 11 reported lower cerebrospinal fluid (CSF) temperature in patients with AD.The association between BT and mini-mental state examination (MMSE) scores has not been investigated in the literature.Thus, there seems to be no consensus on the relationship between BT and cognitive status.The ALPS index also seems to be related to MMSE or age. 2,12,13However, the relationships between BT, ALPS index, and amyloid deposition in the cerebral cortex remain unclear. 14,15Revealing the association between metabolic MRI measurements and amyloid PET would be useful in future studies or clinical situations that evaluate AD.
Therefore, the aim of this study was to comprehensively examine the relationship among BT, ALPS index, and amyloid PET deposition with clinical information such as age, sex, and MMSE in AD patients and healthy controls.

Subjects
Approval from the ethics board of our institution was not required, because we utilized an external open-source data.All data were downloaded from the Open Access Series of Imaging Studies (OASIS-3) brain project (http://oasis-brains.org/), which is a neuroimaging dataset for normal aging and AD and is freely available to the scientific community. 16This dataset did not contain any studies involving human participants or animals conducted by any of the authors within our institution.The dataset contained 2842 MR sessions of 1379 subjects.The inclusion criteria were participants who had 3 T DTI data available in 64 directions and underwent 18 F-florbetapir PET performed on the same day.Subsequently, 415 participants, including 30 patients with AD, were selected.The exclusion criteria were those whose image data was not clear enough to utilize.Among these patients with AD, one had unavailable amyloid PET data; hence, this patient was excluded from the study.Age-and sex-matched 29 normal controls (NCs) were chosen from the 385 remaining normal subjects.Finally, 29 patients with AD and 29 matched NCs were enrolled in the study.Clinical information, including age, sex, and MMSE, was retrieved from the Alzheimer Disease Research Center data from the date closest to that of the MR examination (mean, 176 days; maximum, 361 days).

MRI Preprocessing
All MRI data from the OASIS-3 database were acquired using a 3T Siemens scanner (Biograph mMR, Siemens, Erlangen, Germany).The detailed acquisition parameters are available on the OASIS-3 website. 17Anatomical images were obtained using T1-weighted magnetization-prepared rapid gradient-echo with repetition time (TR)/echo time (TE)/inversion time (TI) = 2300/2.9/900msec, flip angle = 9 , matrix size 240 Â 240, and slice thickness = 1.2 mm without an interslice gap.The DTI parameters consisted of 64equidistant diffusion-sensitizing directions with a b value of 1000 sec/mm 2 , along with a single b value = 0 sec/mm 2 image, with TR/TE = 11,000/87 msec, flip angle = 90, matrix size = 96 Â 96, and slice thickness = 2.5 mm without an interslice gap.Using MRtrix3 software (dwidenoise, mrdegibbs, dwifslpreproc, dwibiascorrect with ants option; https://www.mrtrix.org),all DTI was denoised, Gibbs ringing artifacts were removed, and the motion, eddy current, and bias fields were corrected. 18or the automatic segmentation of the gray matter of the cerebrum, cerebellum, and lateral ventricle, FreeSurfer software was used (recon-all, version 7.2; MGH, Boston MA, USA; https://surfer.nmr.mgh.harvard.edu),and the segmentation was checked by a neuroradiologist with 12 years of experience (H.T.).As no gross segmentation error was observed, no corrections were made.Finally, the gray matter region of interest (ROI) of the cerebrum and cerebellum, as well as the ROI of the lateral ventricles, were created for each subject.The gray matter ROI of the cerebrum included the bank of the superior temporal sulcus, caudal anterior cingulate, corpus callosum, cuneus, fusiform, inferior temporal lobe, isthmus of the cingulate gyrus, lateral occipital lobe, lateral orbitofrontal lobe, lingual lobe, medial orbitofrontal lobe, middle temporal lobe, parahippocampus, paracentral lobe, pars opercularis, pars orbitalis, pars triangularis, pericalcarine, postcentral lobe, posterior cingulate, precentral lobe, precuneus, rostral anterior cingulate, rostral middle frontal lobe, superior frontal lobe, superior parietal lobe, superior temporal lobe, supramarginal lobe, frontal pole, temporal pole, transverse temporal, and insula.The ROIs of the cerebrum and cerebellar cortex were used for the subsequent PET analysis.
To register DTI, images obtained with b = 0 sec/mm 2 were rigidly aligned to the anatomical images and the transform and inverse-transform matrices were calculated using a 6-degree of freedom rigid transformation and a mutual information cost function using the FSL software (flirt, version 6.0; FMRIB, Oxford, UK; http://www.fmrib.ox.ac.uk/fsl/).Inverse transform matrices were applied to the ROI of the lateral ventricles and then the ROI of the lateral ventricle was created in the DTI space for the following BT analysis.[21]

Calculation of BT
The temperature in the lateral ventricular CSF was considered as a representative indicator of brain core temperature. 1,10,11This was based on its direct contact with the brain parenchyma and the unrestricted diffusion of water in this region. 4,22,23The CSF temperature can be estimated using this relationship based on the studies of Mills 22 and Kozak et al. 23 by using the following equation: where D is the diffusion constant (mm 2 /sec); b is the applied diffusion weighting (sec/mm 2 ); and S 0 and S are the voxel signal intensities of the reference (b0 image) and DWIs, respectively.D was converted to the corresponding temperature using the following equation: where T is the temperature ( C).Small diffusion coefficients along the lateral ventricular wall and large diffusion coefficients caused by the CSF flow were excluded based on the thresholds.To establish the thresholds for each subject, we first randomly identified the signal intensities for the gray matter, white matter, and lateral ventricles on the b0 image.Then, we selected a suitable threshold.This threshold enabled us to isolate only voxels within the lateral ventricles that exhibited values exceeding the chosen threshold.The histogram curve of every temperature value within the lateral ventricle was generated in MATLAB (Version R2022a; MathWorks, Natick, Massachusetts, USA) using the curve-fitting method proposed by Sakai et al. 24 Then, the BT value was determined as the most frequently occurring number in the histogram curve.

Calculation of ALPS Index
Color-indexed fractional anisotropy maps were obtained from the aforementioned DTI scans using AFNI (NIMH Scientific and Statistical Computing Core; Bethesda, MD, USA; https://afni.nimh.nih.gov). 25Four 5 mm spherical ROIs were marked by a neuroradiologist (S.M.) with 5 years of experience.Another neuroradiologist (H.T.) with 12 years of experience validated these ROIs in the region of the projection and association fibers for both hemispheres and adjacent to the medullary veins at the level of the lateral ventricle body.Care was taken to ensure that the ROIs were exclusively located in the selected fiber tracts.The DTI-ALPS index measures the ratio of the mean bilateral x-axis diffusivity in the area of projection fibers (D x,proj ) and bilateral x-axis diffusivity in the area of association fibers (D x,assoc ) to the mean bilateral y-axis diffusivity in the area of projection fibers (D y,proj ) and bilateral z-axis diffusivity in the area of association fibers (D z,assoc ).The formula is given by: ALPS index ¼ mean D x,proj , D x,assoc À Á =mean D y,proj , D z,assoc À Á An ALPS index close to 1.0 indicates minimal diffusivity along the perivascular space, while a higher ALPS index indicates greater diffusivity.
Acquisition of 18 F-Florbetapir PET and Calculation of Standardized Uptake Value Ratio (SUVR) Dynamic PET taken for 70 minutes were acquired, starting with the intravenous administration of approximately 370 MBq (110 mCi) of 18 F-florbetapir.The images acquired between 50 and 70 minutes were then summed together to create a static image of amyloid PET.The static images were registered to the anatomical images for each subject using a 6-degree of freedom rigid transformation and a mutual information cost function using FSL software (flirt).Amyloid PET was normalized to create SUVR maps by dividing the values of each voxel by the mean value of the cerebellar cortex.Subsequently, the mean SUVR of the cerebral cortex was calculated.
Preprocessing overview is shown in Fig. 1.

Statistical Analysis
Normality was assessed using the Kolmogorov-Smirnov test.Nonparametric variables were compared between the AD and NC groups using the Mann-Whitney U test, while parametric variables were compared using Welch's t-test.The Pearson's or Spearman's correlation coefficient was computed for each group (AD and NC).A multiple linear regression model was used to assess the association between BT, ALPS index, and SUVR, using age, sex, and group (AD or NC) as covariates.Similarly, multiple linear regression was used to assess the association between BT, ALPS index, and SUVR using age, sex, and MMSE score as covariates.P values <0.05 were defined as statistically significant.Statistical analyses were performed using GraphPad Prism version 9.4.0 software (GraphPad Software, San Diego, CA, USA; https:// www.graphpad.com/scientific-software/prism/).

Results
The characteristics of the subjects are presented in Table 1.Statistically significant differences were found between the AD and NC groups in terms of MMSE, ALPS index, and SUVR, whereas no significant difference was found in BT (P = 0.46).A scatter plot evaluating Pearson's correlation analysis of every pair of BT, ALPS index, and SUVR is shown in Fig. 2. A significant positive correlation was found between BT and ALPS index among NCs (r = 0.44).Figure 3 shows a scatter plot and a linear regression line between the imaging metrics and age.A significant negative correlation was found between age and ALPS index (r s = À0.43 among the AD group and À0.47 among the NC group).
The results of the multivariate linear regression analysis, which aimed to assess the association between BT, ALPS index, and SUVR using age, sex, and group (AD or NC) variables were as follows: age was significantly associated with BT, all three variables were significantly associated with the ALPS index, and group was significantly associated with SUVR (Table 2).
A scatter plot and a linear regression line between the imaging metrics and MMSE, and a multivariate linear regression analysis using MMSE are shown in the Data S1.

Discussion
This study investigated the association between amyloid PET, metabolic brain MRI, and clinical factors using an opensource dataset with 29 AD patients and 29 age-and sexmatched NCs.MMSE, ALPS index, and SUVR showed FIGURE 1: Preprocessing overview of MRI.Anatomical segmentation was performed using T1-weighted magnetization-prepared rapid gradient-echo images, then the gray matter ROI of the cerebrum and cerebellar, as well as the ROI of the lateral ventricles were created for each subject.DTI was registered by aligning b = 0 sec/mm 2 images to the anatomical images e. Lateral ventricle ROI was applied by inverse transform matrices at the DTI space and then BT was calculated.Color-coded and diffusivity maps were created for calculating ALPS index.After administration of 18 F-florbetapir, images between 50 and 70 minutes were summed to create a static amyloid PET image.Subsequently, cerebral cortex ROI was applied to this image to calculate SUVR.ROI, region of interest; DTI, diffusion tensor image; ALPS index, index of diffusivity along the perivascular space; PET, positron emission tomography; SUVR, standardized uptake value ratio.significant differences between the AD and NC groups.A significant positive correlation was found between BT and ALPS index among the NC group, while a significant negative correlation was found between age and ALPS index among the AD and NC groups.In the multivariate linear regression analysis, age was significantly associated with BT, group and   age were significantly associated with ALPS index, and group and MMSE were significantly associated with SUVR.
The observed significant positive correlation between BT and the ALPS index in this study may reflect the association between BT and impairment of the glymphatic system.Since both imaging metrics showed a significant correlation with age in multiple regression analysis, age could be considered a confounding factor.The BT metric reflects the diffusivity of water molecules in the lateral ventricle and may affect the water diffusivity in the perivascular space of the periventricular white matter, which is an adjacent anatomical structure.
Among the three imaging metrics, the pair of SUVR and BT and that of SUVR and ALPS index showed no significant correlation.Amyloid accumulation happens mainly in the cerebral cortex and may be present up to 20 years before the onset of dementia, whereas the onset of changes in BT or ALPS index in AD patients is not known. 26Additionally, Ota et al. 27 have reported negative correlations between ALPS index and SUVR of PiB (Pittsburgh Compound B) only in temporal and parietal cortices.These differences may have influenced the results of the current study.In this study, patients with AD and age-and sex-matched controls were selected from the OASIS-3 dataset; however, the number of patients with AD may have been insufficient.Furthermore, OASIS-3 dataset did not provide the information of mild cognitive impairment patients, including a preclinical or prodromal stage of AD.If more cases are collected, including patients with mild cognitive impairment, along with patients with AD and NCs, it is possible that a correlation may emerge.Contrary, the usefulness of amyloid imaging for the diagnosis and management of patients with AD is a topic of ongoing discussion. 28he current study showed that the group and MMSE scores were significantly associated with SUVR.This was consistent with previous studies, using 18 F-florbetapir as a tracer, [29][30][31] and may have reflected the widely recognized fact that amyloid PET is correlated with the presence and density of beta-amyloid. 29T showed a negative correlation with age, which was consistent with a previous study. 1Our study raises a question about what metabolic process BT represents.As adenosine triphosphate is known to be heavily consumed at the synapses of neurons, a decline in synaptic activity due to aging may be a direct cause of a decline in metabolism.In the literature, the association between BT and AD seems to be controversial potentially due to the limited sample size and diverse disease duration. 10,11BT and MMSE scores were not significantly correlated in the current study.This may be due to the narrow distribution of MMSE scores, especially in NCs, which makes it difficult to find statistically significant differences.
In the current study, age and ALPS index showed a negative correlation not only in NCs but also in AD patients.
Age and presence of AD were significantly associated with the ALPS index in the multivariate regression analysis, which also supported that these two factors were independently associated with the ALPS index.Several reports have shown negative correlations between age and the ALPS index in normal subjects and in patients with Parkinson's disease. 32,33Taoka et al. 13 also indicated a correlation in normal subjects between age and the ALPS index with a slight peak around 40 years.Naganawa et al. 34 reported that the apparent diffusion coefficient of the white matter increases with aging.Beck et al. 35 documented fractional anisotropy of the white matter, showing a linear decline with aging.This indicated impaired water diffusivity of the white matter due to aging, leading to a lower ALPS index as the patient ages.In contrast, the exact effect of decreased ALPS index in AD patients remains unclear as same in aging subjects for now.Dysfunction of the glymphatic system was suggested to cause an imbalance between the production and clearance of amyloid accumulation, reflecting the decreased ALPS index in the AD group, although the current study did not show a direct correlation between the SUVR and ALPS index.The disease duration of AD may also be a confounder, although it is difficult to determine the disease onset in this amyloid deposit disorder, as described above.The ALPS index of the AD group was significantly lower than that of the NC group, while MMSE was not significantly associated with the ALPS index in this study, which is contrary to previous studies. 2,12Compared with the previous study of Taoka et al., 2 the cohort of this study had a wider range of MMSE scores; however, the variance of this cognitive score seems to be different from that of their population, which may cause disagreement.

Limitations
First, the number of AD patients were small due to the strict inclusion criteria.Since BT, ALPS index, and age seems to be intricately related to each other, prospective validation using a large dataset is required in future studies to reveal correlations between metabolic brain MRI, amyloid deposition, and clinical information.Second, the assessment of SUVR in the entire cerebral cortex may have potentially impacted the outcomes of the study.Third, disease duration which may have affected BT and ALPS index was not obtained in the OASIS-3 dataset.Fourth, there could be an interdependence between BT and ALPS index due to their shared dependence on DTI data.Finally, the intervals between the MR examination and retrieved date of clinical information were relatively long.Since neurodegenerative diseases tend to progress slowly, this would not have a significant impact on our study.

Conclusion
This study revealed a positive correlation between BT and ALPS index, which might reflect the association between BT and impairment of the glymphatic system.However, no significant correlations were found between SUVR of amyloid PET and other metabolic MRI metrics.Notably, age and presence of AD were independently associated with the ALPS index.These results will be useful in future research to validate the association between metabolic MRI measurements and amyloid PET.

FIGURE 2 :
FIGURE 2: Scatter plots between (a) BT and ALPS index, (b) BT and SUVR, and (c) ALPS index and SUVR.Each line depicts the linear regression lines of all subjects, AD group, and NC group.*Statistically significant.BT, brain temperature; ALPS index, index of diffusivity along the perivascular space; SUVR, standardized uptake value ratio; AD, Alzheimer's disease; NC, normal control.

FIGURE 3 :
FIGURE 3: Scatter plots between (a) age and BT, (b) age and ALPS index, and (c) age and SUVR.Each line depicts linear regression lines of all subjects, AD group, and NC group.*Statistically significant.BT, brain temperature; ALPS index, index of diffusivity along the perivascular space; SUVR, standardized uptake value ratio; AD, Alzheimer's disease; NC, normal control.

TABLE 2 .
Results of Multiple Linear Regression Analysis a Statistically significant.BT, brain temperature; ALPS index, index of diffusivity along the perivascular space; SUVR, standardized uptake value ratio; AD, Alzheimer's disease; NC, normal control.