Quantitative susceptibility mapping to evaluate the early stage of Alzheimer's disease

The objective of this study was to evaluate susceptibility changes caused by iron accumulation in cognitive normal (CN) elderly, those with amnestic mild cognitive impairment (aMCI), and those with early state AD, and to compare the findings with gray matter volume (GMV) changes caused by neuronal loss. The participants included 19 elderly CN, 19 aMCI, and 19 AD subjects. The voxel-based quantitative susceptibility map (QSM) and GMV in the brain were calculated and the differences of those insides were compared among the three groups. The differences of the QSM data and GMVs among the three groups were investigated by voxel-based and region of interest (ROI)-based comparisons using a one-way analysis of covariance (ANCOVA) test with the gender and age as covariates. Finally, a receiver-operating-characteristic (ROC) curve analysis was performed. The voxel-based results showed that QSM demonstrated more areas with significant difference between the CN and AD groups compared to GMV. GMVs were decreased, but QSM values were increased in aMCI and AD groups compared with the CN group. QSM better differentiated aMCI from CN than GMV in the precuneus and allocortex regions. In the accumulation regions of iron and amyloid β, QSM can be used to differentiate between CN and aMCI groups, indicating a useful an auxiliary imaging for early diagnosis of AD.


Introduction
Alzheimer's disease (AD) is the most common progressive and complex age-related neurodegenerative disorder, leading to cognitive decline (Albertini et al., 2012) and memory loss (Jahn, 2013). Mild cognitive impairment (MCI) is a high-risk condition for the development of clinically probable AD or other neurological conditions (Petersen et al., 2001). The onset of AD is commonly preceded by an interim phase known as amnestic MCI (aMCI) (Shah et al., 2000). The assessment of aMCI is beneficial in terms of early AD intervention and perhaps of AD prevention (Sherwin, 2000). The hallmarks of AD may manifest as alterations of susceptibility in a certain area in the brain, caused by associated iron overload (Haacke et al., 2005), and demyelination of white matter (Carmeli et al., 2014). Loss of myelin causes the susceptibility of white matter to increase, approaching that of gray matter (Lee et al., 2012). Both demyelination and iron deposition increase local tissue susceptibility . Elevated levels of brain iron have been suggested to be a risk factor for AD (Duce et al., 2010;Ayton et al., 2015).
Due to the relevance of susceptibility properties to physiological changes, a number of studies have been performed to observe susceptibility changes using MRI, especially to detect increase iron contents by T2* map (Saarlas et al., 2006) and by susceptibility weighted images (SWI) (Haacke et al., 2009). However, those MRI techniques are not useful for measuring because they suffer from blooming artifacts (Li et al., 2012) and cannot quantify tissue susceptibility (Schenck and Zimmerman, 2004). Quantitative susceptibility mapping (QSM) (de Rochefort et al., 2010;Wang and Liu, 2015) is a MRI technique that enables quantification of susceptibility-changing materials (Schweser et al., 2012).
Magnetic susceptibility refers to a physical property of a material that is useful for chemical identification and quantification of specific materials, including iron, calcium, and contrast agents. Most biological substances, such as calcium and white matter myelin, are slightly diamagnetic, meaning that they induce small negative susceptibility changes. However, iron stored in ferritin, hemosiderin, and neuromelanin in brain tissue and iron embedded in deoxyhemoglobin in venous blood is highly paramagnetic and causes strong magnetic field (Bilgic et al., 2012).
QSM can be used to study iron involvement in AD, but few papers have reported the difference of the QSM values between AD and controls in a very limited subject population (Acosta-Cabronero et al., 2013;Moon et al., 2016). Although previous studies have examined the brain iron concentration in normal aging and reported different susceptibility values between AD and controls using R2* and SWI, no studies have examined the susceptibility changes in the early state of AD using QSM. Furthermore, in vivo iron distributions in brains of patients with early state AD are still unknown. The objective of this study was to systematically investigate the susceptibility changes in subjects of CN, aMCI, and early stage AD using voxel-based analysis and region of interest (ROI)-based analysis, and to compare the data to gray matter volume (GMV) changes caused by neuronal losses in the subjects.

Subjects
Our institutional review board approved this prospective study and informed consent was obtained from all participants. All participants provided a detailed medical history and underwent a neurologic examination, standard neuropsychological testing, and MRI scan. Cognitive functions were assessed using the full Seoul Neuropsychological Screening Battery (SNSB), which is a standardized neuropsychological battery in Korea that comprises tests for attention, visuospatial function, verbal and visual memory functions, languagerelated function, and frontal executive function (Han et al., 2010). A neuroradiologist with 15 years of imaging experience evaluated brain MR images for each subject to determine if study subjects had any evidence of prior cortical infarctions or other space-occupying lesions.
Based on the results of SNSB examination and MRI findings, CN were selected from healthy volunteers who did not have a medical history of neurological disease, who showed normal results on detailed cognitive testing scores that were within 1 standard deviation (SD) adjusted for age, gender, and education according to the Korean normative database, and who also had a normal brain MRI. Patients with mild and probable AD were included, and these patients were defined as those with the Clinical Dementia Rating (CDR) scores of 0.5, 1, or 2, according to the National Institute of Neurological and Communicative Disorders and Stroke-Alzheimer Disease and Related Disorders Association (NINCDS/ADRDA) criteria (McKhann et al., 1984). aMCI was also included, according to the Petersen criteria (Petersen et al., 1999). Patients with vascular dementia were excluded in this study.
We recruited a total of 69 subjects from the neurologic clinic center in our institute. After reviewing clinical database as well as MRI, 12 subjects were excluded because of streaking artifacts in QSM data (n = 5), wrong scanning of the T1-weighted image (n = 1), brain hemorrhage (n = 1), abnormal neuropsychological findings (n = 2), and withdraw prior to finishing the SNSB examination and MRI scan (n = 3). The remaining 57 participants included 19 elderly CN, 19 aMCI, and 19 mild and probable AD. Table 1 summarizes the demographic characteristics and results of neuropsychological tests of the subjects.

Processing of 3D FFE images to obtain QSM
To generate QSM, the acquired magnitude and phase images from the 3D FFE sequence were further processed by implementing the Morphology Enabled Dipole Inversion (MEDI) method (Liu et al., 2012a;Liu et al., 2011b;Liu et al., 2012b) with the following steps: estimation of the total field (Wei Xu, 1999), generation of the tissue field to remove the background field using the projection onto dipole fields (PDF) (Liu et al., 2011a), process of the field-to-susceptibility inversion to solve the ill-posed inverse problem, and generation of QSM map.

Post-processing of QSM and gray matter volume from 3D T1 W image
The following post-processing steps were performed using a Statistical Parametric Mapping Version 8 (SPM8) program (Wellcome Department of Imaging Neuroscience, University College, London, UK). First, the 3D T1W image was co-registered to a magnitude image acquired from the 3D FFE sequence of the same subject. All 3D T1W images were then segmented into gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF). Using these segmented tissues, the study-specific brain template was created using a Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra (DARTEL)  Note. The data of age and K-MMSE scores are presented as the mean ± standard deviation, but those of CDR scores are presented as the median (range) value. a Age in years significantly differed between the CN and other groups (p < 0.05), but there was no significant difference between the aMCI and AD group (p > 0.05). b Genders are significantly different between CN and aMCI (p = 0.0013), but not between aMCI and AD (p = 0.0665) and also between CN and AD (p = 0.0665). c K-MMSE scores were significantly different between the CN and other groups (p < 0.05) and between the aMCI and AD groups (p < 0.05).
tool (Ashburner, 2007). The segmented GM of each subject was spatially normalized into the created template. The spatially normalized GMV was smoothed using the Gaussian kernel of 8 × 8 × 8 mm full width at half-maximum (FWHM).
Second, the QSM map of each individual subject was spatially normalized into the created brain template on the MNI space using the normalized parameters of the 3D T1W image. After the spatial normalization, the QSM maps were processed to minimize contamination  GMV (A) obtained from a 3D T1-weighted imaging was higher in CN than that in AD, but QSM value (B) was higher in AD than that in CN. There were no significant differences for both GMVs and QSM values between other subject groups.
of the CSF effects, which was not a concern for the study, by using the following steps. Voxels containing > 80% GM and WM obtained from the segmented 3D T1W images were selected and were masked out otherwise. The mean QSM value in each voxel was subtracted by the internal standard reference values which were estimated by the average of the QSM values in the bilateral posterior ventricular region for each subject (Acosta-Cabronero et al., 2013), since the values calculated from MEDI are relative with an unknown offset. Finally, QSM maps were smoothed by using the Gaussian kernel of 8 × 8 × 8 mm FWHM. Note. For both QSM and GMV data, there were significant differences between the CN and AD groups, but no significant difference was evident between the other groups. The result of statistically significant differences of GMV and QSM are indicated by circle (○) for the comparison between CN and aMCI groups, of plus (+) for the comparison between aMCI and AD group, and of asterisk (*) for the comparison between CN and AD group. GMV was lower in aMCI and AD than that in CN, but QSM value was higher in aMCI and AD than in CN.
2.6. Statistical analyses 2.6.1. Demographic characteristics and results of neuropsychological tests Demographical data and clinical outcome scores were compared among the three groups. Age and Mini-Mental State Examination (MMSE) scores were not normally distributed (p < 0.005 by Kolmogorov-Smirnov test). Hence, age and MMSE scores were tested using the Kruskal-Wallis test. If we found any significant differences among the groups, then we performed a post hoc test for pairwise comparison of subgroups according to the Conover method (Conover, 1999). Gender was tested using the Chi-squared test.

Voxel-based analysis of GMV and QSM
To compare GMV and QSM among the three groups, voxel-wise, one-way analysis of covariance (ANCOVA) test was used with subject's age and gender as covariates. A significance level of p = 0.05 was applied with correction for multiple comparisons using the family-wise error (FWE) method and clusters with at least 10 contiguous voxels.

ROI-based analyses of GMV and QSM
Because ROI for GMVs and QSM were normally distributed (p > 0.05 by the Kolmogorov-Smirnov test), we used the following parametric test. Because we had several different ROIs within each subject and the three subject groups, two-factor repeated measures ANOVA was used as the first factor for the three subject groups. The second factor was ROIs. In this analysis, we evaluated the betweensubject effects and the within-subject effects. After the two-factor analysis, we reanalyzed GMVs and QSM using the one-factor repeated measures ANOVA to evaluate differences of GMVs and QSM among ROIs. Furthermore, to evaluate group differences of GMVs and QSM among the three subject groups for each ROI, ANCOVA was used with age and gender as covariates. The post-hoc test was performed by using the pairwise comparisons of those values between subject groups with Bonferroni corrected p = 0.05.

Receiver operating characteristic (ROC) curve analyses
ROC curve analysis was performed to demonstrate sensitivity and specificity of GMVs and QSM to differentiate among the subject groups for each ROI. A significance level of Bonferroni corrected p = 0.05 was also applied for this analysis. MedCalc statistical software (www. medcalc.org, Ostend, Belgium) was used to analyze the ROI. Table 3 Mean values for GMVs and QSM values obtained from the specific region-of-interests (ROIs) defined in the seven well-known iron accumulation regions and the five amyloid β accumulation regions in the subjects with Alzheimer's disease (AD), amnestic mild cognitive impairment (aMCI), and cognitively normal (CN) elderly.

ROI
Group ( Note. The result of statistically significant differences of GMV and QSM are indicated with the symbols of circle (○) for the comparison between CN and aMCI groups, of plus (+) for the comparison between aMCI and AD group and of asterisk (*) for the comparison between CN and AD groups. The significance for data with italics emphasis is Bonferroni corrected p = 0.05.

Subject characteristics
The demographic data of the subjects are summarized in Table 1. Ages were significantly different among the three groups (p = 0.0125). Ages were significantly different between the CN and other groups (p < 0.05) and between the aMCI and AD (p < 0.05). Genders were significantly different between the CN and aMCI (p = 0.0013), but not between the aMCI and AD (p = 0.0665) and also between the CN and AD (p = 0.0665). The MMSE scores were significantly different between the groups (p < 0.0001). MMSE scores were significantly different between the CN and other groups (p < 0.05) and between the aMCI and AD (p < 0.05). Fig. 1 displays representative images of GMVs (Fig. 1A, C, and E) and the corresponding QSM (Fig. 1B, D, and F) for the CN (Fig. 1A and B), aMCI ( Fig. 1C and D) and AD ( Fig. 1E and F). QSM maps are shown with voxels combining > 80% of GM and WM.

Voxel-based analysis
The results of voxel-based comparisons of GMV and QSM among the three groups are summarized in Fig. 2 and Table 2. QSM and GMV were significantly different between the CN and AD, but no significant difference between other groups. Increased QSM values in AD compared with CN were found mainly at the left precentral gyrus, left postcentral gyrus, right caudate, right cingulate gyrus, right parahippocampal gyrus, hippocampus, amygdala, caudate, and insula. Decreased GMV in AD compared with CN was found at the left and right thalamus, and the left anterior cingulate, hippocampus, caudate, and cingulate gyrus. Fig. 3 shows the percent changes of GMV (Fig. 3A) and QSM (Fig. 3B) values for aMCI and AD compared to CN subjects for each ROI. GMV were significantly different between-subject groups (DF = 2, F = 13.57, p < 0.001) and within-subject group (DF = 12, F = 5446.36, p < 0.0001). QSM were significantly different betweensubject groups (DF = 2, F = 5.63, p = 0.006) and within-subject group Table 4 Results of ROC curves analysis of GMVs and QSM values obtained from the seven well-known iron accumulation regions-of-interests (ROIs) in the subjects with Alzheimer's disease (AD), amnestic mild cognitive impairment (aMCI), and cognitively normal (CN) elderly. Abbreviations: ROC, receiver operating characteristic; AUC, area under the ROC curve; GMV, gray matter volume; QSM, quantitative susceptibility mapping; SE, sensitivity; SP, specificity; p, p-value.

ROI
The significance for data with italics emphasis is Bonferroni corrected p = 0.05.
(DF = 12, F = 452.87, p < 0.001). These results indicated that GMV and QSM values were significantly different among the 13 ROIs as well as were significantly different among the three groups. We describe the results of the post-hoc test of GMV and QSM values below.

GMV
Compared to the CN, GMV in the aMCI was significantly decreased at the hippocampus (p = 0.0208). GMV in the AD was significantly decreased at hippocampus (p < 0.0001), amygdala (p = 0.0394), and putamen (p = 0.0077). Compared to the aMCI, GMV in the AD was significantly decreased only at the hippocampus (p = 0.0209) ( Table 3).
3.3.3.2. Five amyloid β accumulation ROIs. ROC curves analysis of GMVs and QSM values obtained from the five amyloid β accumulation ROIs in the subjects with CN, aMCI, and AD are summarized in Table 5  differentiated between CN and AD in the neocortex (AUC = 0.734, p = 0.0055). Fig. 4D shows that the area under curve (AUC) values in the allocortex were greater in QSM (0.7590, p = 0.0011) than those in GMVs (0.5373, p = 0.6997) to differentiate aMCI from CN.

Relationship between susceptibility effects and iron accumulation regions in the brain
QSM values were significantly increased from CN to AD in the hippocampus, amygdala, precuneus, and thalamus, indicating that the susceptibility-induced contents accumulated with disease progression. QSM values also differentiated aMCI from CN (Table 4). The precuneus was most sensitive to monitoring early changes. Measurement of GMV changes can be also used to differentiate groups. The hippocampus and amygdala differentiated the groups with QSM and GMV. Iron levels at the cellular level are reported to be significantly elevated in neurofibrillary tangle using cytologic or immunocytochemical methods (Moretz et al., 1990). Our study is consistent with the suggestion of an iron overload in AD and supports the concept that iron homeostasis is disrupted in the AD brain. Presently, regional changes in iron accumulation were evident in brains of AD, except for the putamen. This means that in vivo measurement using QSM be a diagnostic tool for AD. It is still unclear whether iron accumulation is the cause or a consequence of the neurodegenerative cascade. For the present, this study provides important insights for the understanding of AD pathogenesis by monitoring the spatial distributions of iron deposition in cognitively normal and impaired individuals. Although QSM cannot distinguish which iron contribute to signals, the accumulation of iron from CN to AD in specific brain regions could help clarify the understanding of AD pathogenesis.

Relationship between susceptibility effects and the amyloid β protein in the brain
Brian regions displaying amyloid β accumulation in PET were significant susceptibility changes in AD, but GMV did not change (Table 3, Fig. 3). Furthermore, QSM values at the allocortex, ACC, and PCC were depicted in early changes of susceptibility in aMCI. However, GMV were not significantly different among the three groups in the predefined ROIs. Therefore, QSM may be much more sensitive than GMV to investigate group differences at those regions. While QSM values obtained from the neocortex, allocortex, ACC, and PCC were differentiated between CN and aMCI, GMV was not. This result may be explained by the early accumulation of amyloid β at those areas in the early stage of diseases without changing neuronal degeneration. This result explains that the amyloid imaging by PET is associated with susceptibility mapping by MRI.
One important feature of AD is associated with depositions of amyloid β peptide (Cuajungco et al., 2000;Hardy and Selkoe, 2002;Vandenberghe et al., 2013). The amyloid plaque overloads aggregate irons in brain tissue to induce chemical reduction of redox-inactive ferric irons to redox-active ferrous iron forms, producing oxidative stress and resultant neuronal damages (Everett et al., 2014). Because increased iron is closely related to greater production of amyloid β peptides, the assessment of brain iron levels has been a major hallmark of AD (Raven et al., 2013). One of the causes of the increased susceptibility value on AD tissue may be related to elevate iron and ironmediated redox activity, even at preclinical and prodromal stages of AD Abbreviations: ACC, anterior cingulate cortex; PCC, posterior cingulate cortex; ROC, receiver operating characteristic; AUC, area under the ROC curve; GMV, gray matter volume; QSM, quantitative susceptibility mapping; SE, sensitivity; SP, specificity; p, p-value.
The significance for data with italics emphasis is Bonferroni corrected p = 0.05. (Smith et al., 2010). Neurofibrillary tangles develop beginning in the neocortical area (phase 1) and then spread to allocortical regions, such as the entorhinal cortex, and the anterior and posterior cingulate gyrus (phase 2) (Thal et al., 2002). Therefore, our results of QSM changes in these cortical lesions in AD and aMCI may indicate the contributions of both amyloid accumulation and neurofibrillary tangles. Presently, the susceptibility alterations were revealed to already start in the aMCI stage, which can be used as the early imaging marker. The QSM values can be distinguished CN from AD, as well as CN from aMCI groups. GMV may be classified between the CN and AD, but not between the CN and aMCI. Therefore, the QSM values in the precuneus and allocortex are more effective to investigate the early changes in the brain than GMV values. The quality of QSM data is depended on the optimization for the sequence parameters and post-processing algorithm. A clinical setting of QSM requires to shorten the scan time, reduce any artifacts, and to easily implement quantifications of the susceptibility. QSM still has a limitation to use for the individual clinical case to diagnosis iron uptake in a patient, indicating that QSM has to further develop to directly use as an imaging biomarker in individual patients.

Conclusions
The susceptibility difference in cognitive normal (CN), amnestic mild cognitive impairment (aMCI) and Alzheimer's disease (AD) elderly was more sensitive than gray matter volume (GMV) change in known regions of iron and the amyloid β accumulations, indicating the susceptibility changes caused by iron accumulation in the brain of Alzheimer's disease. Especially, QSM was better to differentiate between the CN and aMCI than GMV. Therefore, the QSM technology can be used as an auxiliary imaging factor for early diagnosis of AD. QSM can be used to evaluate the correlation with amyloid PET diagnosis.