Comparison of MR Perfusion and FDG-PET Brain Studies in Patients With Alzheimer’s Disease and Amnestic Mild Cognitive Impairment

Background: The aim of this study was to compare Dynamic Susceptibility Contrast Enhanced MRI (DSC-MRI) and PET with urodeoxyglucose (FDG-PET) in the diagnosis of Alzheimer’s Disease (AD) and amnestic Mild Cognitive Impairment (aMCI). Methods: Age and sex matched 27 patients with AD, 39 with aMCI and 16 controls underwent brain DSC-MRI followed by FDG-PET. Values of relative Cerebral Blood Volume (rCBV) and rCBV z-scores from frontal, temporal, parietal and PCG cortices were correlated with the rate of glucose metabolism from PET. Sensitivity, specicity and accuracy of DSC-MRI and FDG-PET in the diagnosis of AD and aMCI were assessed and compared. Results: In AD hypoperfusion was found within all examined locations, while in aMCI in both parietal and temporal cortices and left PCG. FDG-PET showed the greatest hypometabolism in parietal, temporal and left PCG regions in both AD and aMCI. FDG-PET was more accurate in distinguishing aMCI from controls than DSC-MRI. In AD and combined group (AD + aMCI ) there were numerous correlations between DSC-MRI and FDG-PET results. Conclusions: In AD the patterns of hypoperfusion and glucose hypometabolism are similar thus DSC-MRI may be a competitive method to FDG-PET. FDG-PET is a more accurate method in the diagnosis of aMCI. the assessment of correlation between the results of DSC perfusion or FDG-PET and the severity of cognitive impairment in AD and aMCI. Our study comparing DSC MR perfusion with FDG-PET results in AD and aMCI lls the gap in the existing scientic literature. temporal and left PCG regions, followed by hypometabolism in the frontal cortices while MCI subjects showed less severe hypometabolism mainly in the parieto-temporal regions and left PCG. Both these results are in accordance with the commonly accepted metabolic pattern in the course of AD [18, 19, 20, 21].


Introduction
Due to aging of the population early diagnosis of dementia is an important problem in modern medicine. The most common cause of dementia is Alzheimer's disease (AD), while amnestic mild cognitive impairment (aMCI) is considered a prodromal condition with a high risk of conversion to AD [1]. Many studies show that brain alterations in AD occur many years before the rst clinical manifestations [2].
Structural magnetic resonance imaging (MRI) plays an important role in the diagnosis of dementia rstly in excluding secondary causes of cognitive impairment such as for example vascular lesions, brain tumors or hydrocephalus, secondly in the assessment of a distribution of brain atrophy. A typical pattern of brain atrophy in the course of AD degeneration involves medial temporal lobes and temporo-parietal areas including posterior cingulate gyrus (PCG), followed by frontal lobe atrophy in advanced cases [3,4]. Moreover, modern advanced MR techniques allow for the assessment of not only brain structure but also its function or metabolism. One of such methods is Dynamic Susceptibility Contrast Enhanced MRI (DSC MRI) enabling insight into the cerebral microcirculation on the basis of evaluation of the rst pass of a contrast material through the brain microvasculature. Dynamic Susceptibility Contrast Enhanced MRI requires administration of contrast material and its results are parametric maps for several perfusion parameters among which Cerebral Blood Volume (CBV) is the most important. Perfusion is the fundamental biological function whereby blood ow is responsible for supplying cells with oxygen and nutrients, for this reason perfusion parameters may anticipate in time structural changes seen later on conventional MRI. Reports on DSC MRI studies in AD patients have shown a signi cant reduction in bilateral CBV values in the temporal-parietal cortex including PCG with a relative sparing of the sensorimotor cortex [5,6,7,8,9] while in aMCI hypoperfusion was reported mainly in PCG [10,11]. Some studies have shown a signi cant correlation of perfusion results with neuropsychological tests in AD and MCI [9,10].
It has to be stressed that in the pathogenesis of AD degeneration, in addition to the amyloid cascade hypothesis, the vascular hypothesis has been postulated recently. On one hand, it is assumed that amyloid itself shows both neuronal and endothelial toxicity (ABSENT hypothesis) leading to brain degeneration and hypoperfusion [12]. On the other hand, cerebrovascular risk factors including higher level of high-sensitivity C-Reactive Protein (hsCRP) and lower level of high-density lipoprotein (HDL) may cause disturbances of macro-or microvasculature circulation and endothelial damage which contributes to amyloid accumulation and to neuronal death [13,14]. In line with the new concept of the neurovascular unit, not only amyloid is angiotoxic, but chronic hypoperfusion and vascular damage further accumulate beta amyloid and exacerbate brain degeneration [15].
Fluorodeoxyglucose positron emission tomography (FDG-PET) is a nuclear medicine technique which looks at the cerebral metabolic rate of glucose (CMR glc) and it has been widely used in the diagnosis of dementia. In AD this examination shows glucose hypometabolism in very speci c locations called "AD metabolic pattern" including temporo-parietal associative cortex, PCG, precuneus, medial temporal lobes, especially in the entorhinal cortex and the hippocampus [16, 17,18,19,20,21]. In the advanced course of the disease changes occur in the frontal cortex, with saving primary sensorimotor cortex. However, in MCI, decreased metabolism is mainly found in PCG, and to a lesser extent in the temporo-parietal area, which may be a sensitive prognostic indicator of conversion to AD [20,22,23,24,25]. Even though FDG-PET is a great method of evaluation of early changes in the brain of AD patients or even in predementia states such as aMCI, its use in everyday clinical practice is limited due to high cost and worse availability. Moreover, it requires injection of radionuclide tracer and uses ionizing radiation since it is performed in conjunction with CT (PET/CT scanner).
Looking at previous studies, structural, perfusion and metabolic changes in AD or aMCI seem to follow the same pattern but there are not many reports comparing them on the same groups of patients. Perfusion and metabolic changes have been reported to proceed structural atrophy but there have been only few reports focusing on direct comparison of MR perfusion with the results of FDG-PET studies in AD and aMCI [5,8,9,11]. Recently, more and more reports are comparing FDG-PET with Arterial Spin Labelling (ASL) MR perfusion which is a technique that does not require injection of any contrast material [26,27,28,29,30]. The ASL results have been found to correlate with SPECT and PET studies [13,26,27,31,32]. Despite the great advantage of the lack of contrast agent, prolonged acquisition time in ASL makes it impossible to use in non-cooperative patients (e.g. with advanced dementia). Next disadvantage of ASL is a low signal-to-noise ratio (SNR) and necessity of scanning with 3 Tesla MR machines [33,34].
Due to low availability of FDG-PET and ASL perfusion we believe that DSC MRI could still play an important role in the diagnosis of dementia as an alternative to those studies. During all the performed MR studies we have not encountered any problems in cooperation with the elderly patients nor any post-contrast side-effects have been reported. It has to be emphasized that modern gadolinium-based contrast materials used in MRI are safe and may cause health problems mainly in patients with severe renal insu ciency.
The major aim of our study was to establish the role of DCS perfusion in relation to FDG-PET imaging based on a detailed comparison of these two techniques. The main assumption of our research was that DSC MR perfusion results should be similar to FDG-PET studies because glucose metabolism is partially dependent on cerebral perfusion. The comparison of DSC MR perfusion and FDG-PET was performed based on: 1) the assessment of hypoperfusion and hypometabolism patterns in the selected brain areas in AD and aMCI, 2) the rate of correlation between the results of these two techniques and their accuracy in diagnosis of AD and aMCI, 3) the assessment of correlation between the results of DSC perfusion or FDG-PET and the severity of cognitive impairment in AD and aMCI. Our study comparing DSC MR perfusion with FDG-PET results in AD and aMCI lls the gap in the existing scienti c literature.

Subjects
The research material consisted of 66 patients: 27 with AD (mean age 70. 33

Magnetic resonance examination
All MR examinations of the brain were performed with a 1.5 Tesla MR scanner (Signa Hdx, GE Medical Systems) using a 16-channel HNS (head-neck-spine) coil. Standard structural protocol was followed by DSC MR perfusion using fast echo planar (EPI) gradient T2*-weighted sequences with the parameters: TR = 1.900 ms, TE = 80 ms, FOV= 30 cm, matrix = 192×128, slice thickness = 8mm without spacing, NEX -1.0. Ten seconds after the start of image acquisition, a bolus of 1.0 mol/l gadobutrol formula (Gadovist, Schering, Berlin, Germany) in a dose of 0.2 ml/kg of body weight was injected via a 20-gauge catheter placed in the antecubital vein. Contrast administration was performed using an automatic injector (Medrad) at a rate of 5 ml/s and was followed by a saline bolus (20 ml at 5 ml/s). The whole perfusion imaging lasted 1 min 26 s in which sets of images from 13 axial slices were obtained before, during, and after contrast injection. The dynamic images were postprocessed into parametric perfusion maps using Functool software (GE, ADW 4.6). Maps of Cerebral Blood Volume (CBV) were computed on a pixel-wise basis from the rst-pass data from the capillary bed. Values of CBV were obtained using manually drawn Regions of Interest (ROIs) within the frontal, temporal and parietal cerebral cortex (500 and 900 mm 2 in size), and the cingulate gyrus (100-200 mm² in size). The CBV values obtained from several ROIs were mathematically averaged to one frontal, temporal or parietal cortical value, separately for the right and left hemisphere. All CBV values were normalized to the mean CBV value of the cerebellar cortex in order to obtain the relative CBV (rCBV). The cerebellar cortex was chosen as the reference area because it is the region less affected in AD compared to other cortical measures [20]. The ROI in the cerebellum was approximately 300-400 mm² in size (Fig. 1). The location of the ROIs was chosen to best correspond with the glucose metabolism measurements in the FDG-PET study (Tab. 2).

PET examination
PET studies were performed within 3 weeks after the MR examination. The PET images were obtained using a GE Discovery STE16 PET/CT scanner with [ 18 F] Fluorodeoxyglucose (FDG) as a radiotracer. All participants fasted at least 6 h before examination. Data acquisition lasted 8 min and was performed 30 min after intravenous injection of 5 MBq/kg of FDG. Detector spatial resolution was 5.6 mm and data were displayed on a 128×128 pixel matrix. To avoid external stimulation during FDG uptake patients stayed in a resting condition in a darkened room. The acquired data were processed using iterative reconstructions. Attenuation and scatter corrections were made simultaneously by transmission measurements using CT. Next, PET/CT images were transferred to a workstation (GE Healthcare) and processed using commercial CORTEX ID application. Scans were spatially normalized to a stereotactic space based on the Talairach and Tournoux atlas [35].
Then brain images underwent size correction to standard dimensions of 3D-atlas and a regional anatomic variants correction to decrease individual variations. All data were normalized to the mean FDG uptake value of the cerebellum, where glucose utilization is comparatively preserved in dementia [36]. Realigned FDG-PET scans of all subjects were compared with a normative, age strati ed reference database included in the CORTEX ID program. Metabolic activity was automatically determined in 14 cerebral regions and z-scores ([mean subjectmean database]/SD database) were calculated. Z-scores data were exhibited as 3D-SSP (Three Dimensional Stereotactic Surface Projection) images [37] to visualize abnormalities with high z-score values pointing out reduction of FDG uptake and glucose hypometabolism (Fig. 2).
Color-coded maps of absolute glucose metabolism -top two rows; color-coded maps of glucose metabolism in z-scores -bottom two rows; parametric values of z-scores of glucose metabolism normalized to the cerebellum -table on the right.

Statistical analysis
To compare DSC MRI with the FDG-PET studies, two types of perfusion parameters were used such as rCBV values and rCBV z-score. The rCBV z-score was used to make the MR results as similar as possible to the FDG-PET results, in which the level of glucose metabolism is automatically presented in the form of z-score that indicates the number of standard deviations of a given parameter from a population norm. A typical formula was used to calculate the rCBV z-score such as (µ-x) / σ with µ representing mean rCBV value for CG, x meaning the mean rCBV value for a given patient and σ standing for standard deviation of CG. In both DSC MRI and FDG-PET higher z-scores meant higher rate of perfusion or metabolic impairment. The comparison of mean age and the results of DSC MRI and FDG-PET between the AD, MCI and CG groups was carried out using the ANOVA method followed by a Scheffe's post hoc test to compare the results in pairs between MCI and CG, AD and CG as well as AD and MCI. In turn, analyzes of correlation between MR perfusion and FDG-PET results as well as between the results of MR perfusion or FDG-PET and the results of psychological tests were performed using the Pearson correlation coe cient. Additionally, the sensitivity and speci city of MR and PET parameters in differentiating between AD, MCI and CG were calculated using the Receiver-Operating Characteristic (ROC) method, in which the accuracy of the test is indicated by the area under the ROC curve. In all statistical analyzes, p value of <0.05 was considered statistically signi cant. In the case of the rCBV z-score and PET z-score parameters, z-scores ≥ 1 were considered to be signi cantly different from the CG.

Results
In AD patients compared to CG, DSC-MRI results showed signi cantly decreased rCBV values (p ≤ 0.05) and signi cantly higher z-core rCBV values (z-score ≥1) in all examined cortical locations.
Compared to healthy controls, MCI patients showed a signi cant decrease of rCBV values within the cortex of both parietal and temporal lobes and left PCG, while using the rCBV z-score signi cant hypoperfusion was found within the right parietal lobe.
The AD group compared to the MCI group, showed signi cantly lower rCBV values and higher rCBV z-scores within all examined areas of the brain cortex (Table 3).  The FDG-PET study in the AD group showed signi cant glucose hypometabolism within all measured areas of the cerebral cortex, while in the MCI group within the cortex of both parietal and temporal regions and left PCG. The greatest impairment of glucose metabolism in patients with AD, as well as in patients with MCI, was demonstrated in the parietal, temporal and left PCG regions. The AD patients compared to MCI subjects showed signi cantly higher impairment of glucose metabolism in all evaluated locations (Table 3).
In AD patients statistically signi cant positive correlations between MR perfusion and FDG-PET results were found almost for all evaluated cortical regions apart from right parietal cortex. In the MCI group there was only one single correlation between these two techniques found within the left PCG (r = 0.4, p = 0.01). In the combined group (AD + MCI) the PET z-score and rCBV z-score analysis showed statistically signi cant positive correlations in all locations. These correlations were strongly positive in the area of PCG and in the temporal lobes (r > 0.5), moderately positive in the area of the parietal lobes, and weaker in other locations (r < 0.5) (Tab. 4). Table 4 The results of correlation between rCBV z-score and FDG-PET z-score separately for AD and MCI groups and for all patients (AD + MCI). In distinguishing MCI from CG the highest sensitivity, speci city and accuracy (0.95, 1.0 and 0.95, respectively) were found for PET z-score, followed by rCBV z-score and rCBV ( Table 5). The highest sensitivity, speci city and accuracy (0.98, 1.0 and 0.98, respectively) in distinguishing AD from CG were revealed for rCBV z-score, followed by FDG-PET z-score and rCBV (Table 5). And lastly, in differentiating AD from MCI the same sensitivity, speci city and accuracy (0.66, 0.94 and 0.84, respectively) were found for both rCBV z-score and rCBV (Table  5).  (Table 6).

Discussion
The aim of our study was to compare DSC MR perfusion and FDG-PET studies based on 1) the assessment of hypoperfusion and hypometabolism patterns in the selected brain areas in AD and MCI, 2) the rate of correlation between the results of these two techniques and their accuracy in diagnosis of AD and MCI, and 3) the assessment of correlation between the results of DSC MRI and FDG-PET with the severity of cognitive impairment in AD and MCI.
In our study AD patients, compared to the control group, showed signi cant hypoperfusion in all examined cortical localizations. Our results are consistent with the typical pattern of Alzheimer's degeneration and hypoperfusion reported in numerous publications within PCG, temporoparietal cortices, and in later stages also frontal cortices with relative sparing of sensorimotor cortex [5,6,9,38,39]. In the MCI group, compared to controls, we found signi cantly decreased rCBV values within the cortex of both parietal lobes, temporal lobes and left PCG, while using the rCBV z-scores signi cant hypoperfusion was detected in the right parietal cortex, which is also consistent with the pattern of very early alterations in the course of AD pathology [6, 10,11,39,40]. In the MCI group perfusion alterations were less severe than in the AD group, which con rms the theory that hypoperfusion is a marker of neuronal damage and becomes more prominent in the later stages of AD.
In our study the FDG-PET results in AD showed the greatest impairment of glucose metabolism in the parietal, temporal and left PCG regions, followed by hypometabolism in the frontal cortices while MCI subjects showed less severe hypometabolism mainly in the parieto-temporal regions and left PCG. Both these results are in accordance with the commonly accepted metabolic pattern in the course of AD [18, 19,20,21].
In our study in the AD patients we found signi cant correlations between the results of DSC MRI and FDG-PET in almost all evaluated locations apart from the right parietal cortex while in the MCI group there was only single correlation within the left PCG. Single correlation in case of MCI was probably due to a small sample of subjects. After combing AD and MCI subjects in one group signi cant correlations between MR perfusion and FDG-PET studies were revealed in all evaluated locations, probably due to the strong in uence of the AD group.
The strongest correlations were revealed within temporal (r = 0.55-0.6) and PCG (r = 0.53-0.63) regions followed by parietal (r = 0. 45 drawbacks. One of them is a prolonged acquisition time which makes ASL impossible to be used in non-cooperative patients (e.g. with advanced dementia). Other disadvantages are the necessity of 3 Tesla MR scanners to obtain reliable data which are not widely available and a low signal-to-noise ratio (SNR).
In the next part of the study we evaluated sensitivity and speci city as well as accuracy of DSC-MRI and FDG-PET studies in distinguishing AD and MCI from healthy controls. We found very similar high accuracy of DSC-MR perfusion and FDG-PET in distinguishing AD from the control group (0.98 and 0.97, respectively), and markedly higher accuracy of FDG-PET than DSC-MR perfusion in differentiation of MCI from the control group (0.96 and 0.68-0.77, respectively lobes, which is consistent with the results of our study where accuracy from this cortical location was greater (0.9-0.92) than in other regions [46]. Regarding differentiation of MCI from CG using DSC-MRI our study showed better results to some previous reports for example by Zimny et al who based on evaluation of PCG determined sensitivity, speci city and accuracy 0.72, 0.8, 0.7 respectively and in the next study accuracy as 0.67. [10,11] [9,10]. The lack of correlation of psychological tests in separate groups of AD and MCI with FDG-PET results is in contradiction with several literature reports [45,47,48,49].
There are a few limitations of our study. Firstly, manual determination of ROIs is somewhat subjective and makes the method operatordependent. Secondly, rather small groups of subjects may have had an impact on some results. We assessed more signi cant correlations after combing patients in a larger group of AD and MCI subjects. Another drawback is a cross-sectional character of the study. We have not evaluated longitudinal results regarding follow-up studies of aMCI subjects and a rate of their progression to dementia. It would be very interesting to check if DSC-MR perfusion has a similar strength as FDG-PET in predicting such a conversion.

Conclusion
In our study we proved that aMCI and AD patients show very similar patterns of hypoperfusion in DSC-MR and glucose hypometabolism in FDG-PET with a high rate of signi cant correlations between these two techniques. FDG-PET seems to a better method in diagnosis of MCI while DSC-MR perfusion was found to be more accurate in diagnosis of AD.
We believe that DSC-MR may be a good alternative to FDG-PET studies in patients with dementia. FDG-PET studies are still not widely available and very expensive while MR examination is a routine study in the work-up of patients with dementia or MCI. A standard MR examination may be easily extended with DSC perfusion which is a fast and fairly easy sequence to be performed.  An example of the FDG-PET study result of a single patient presented as color-coded maps.