Brain iron accumulation in Alzheimer’s disease evaluated using susceptibility-weighted imaging
Amir Fazlollahi1,2, Pierrick Bourgeat1, Ashley I. Bush3,4, Fabrice Meriaudeau 2, David Ames4, Colin L. Masters3,4, Christopher C. Rowe4,5, Victor L. Villemagne3,4,5, and Olivier Salvado1

1Australian e-Health Research centre, CSIRO, Brisbane, Australia, 2University of Burgundy, Le Creusot, France, 3Florey Institute of Neuroscience & Mental Health, Melbourne, Australia, 4The University of Melbourne, Melbourne, Australia, 5Austin Health, Melbourne, Australia


MRI susceptibility weighted imaging (SWI) has shown a promising sensitivity in visualizing iron deposits, while less effort is made to establish a pseudo-quantitative estimate of iron. In this study, an image processing framework was employed to normalize the uncalibrated intensity of SWI with respect to the corresponding value of cerebrospinal fluid. After excluding large detectable veins, the resulting pseudo-quantitative image along with a standard brain atlas, were used to compute regional concentrations of iron in a cohort of Alzheimer’s disease. A group-wise analysis was then showed a stepwise increment in SWI-iron along the progression of the disease.


Increase in iron level has been reported in Alzheimer’s disease (AD) 1,2, potentially receiving contributions from tau in tangles 1 and amyloid-β (Aβ) plaques 3. In recent years, contribution of iron imaging using MRI is rapidly growing due its non-invasive nature and high spatial resolution. Among MRI modalities, susceptibility weighted imaging (SWI) has shown a promising reliability in visualizing iron deposits when compared to T2 or T2*-weighted images 4. However, the main shortcomings of measuring parenchymal iron with SWI are its qualitative measure and its sensitivity to both vascular iron (deoxygenated iron in vessels) and parenchymal iron (ferritin and hemosiderin). In the context of iron changes in AD, previous SWI studies were performed only on subjects with mild cognitive impairment (MCI) and reported non-overlapping findings with the literature 5,6.In this study, an image processing pipeline was utilized to remove the presence of vascular iron (iron in blood vessels) and acquire a pseudo-quantitative iron estimate. A group-wise analysis was then performed to investigate the usefulness of SWI in identifying regional brain iron accumulation associated with AD.


For this study, a subset of 120 cross-sectional scans comprised of 23 AD, 30 MCI and 67 healthy controls (HC) from the Australian Imaging Biomarkers and Lifestyle (AIBL) study were included (for detailed demographic information refer to Table 1).All subjects underwent an anatomical T1-weighted (T1W) and SWI. MRI data were acquired on a 3T Siemens TRIO scanner with a 12-channel head coil, where SWIs were automatically reconstructed online using the scanner system (software VB17). The 3D SWIs were obtained with 0.93×0.93 mm in-plane resolution and 1.75 mm slice thickness, repetition time/echo time of 27/20 msec, and flip angle 20°. The T1W data were parcellated into 45 gray-matter (GM) regions by segmentation propagation of an atlas database which had been previously parcellated using the Automated Anatomical Labeling (AAL) 7. Subsequently, a rigid alignment between the T1W and SWI images was used to propagate the parcellated regions.Given the uncalibrated nature of SWI and lack of specificity to parenchymal iron, the following steps were considered to acquire a pseudo-quantitative estimate of iron: (i) the non-uniform intensity variations caused by magnetic field inhomogeneities were corrected using the N4 bias field correction 8, (ii) the intensity of the bias-free SWI was scaled with respect to the mean value of cerebrospinal fluid (CSF) computed from the lateral ventricles, and (iii) a vessel mask was generated (by performing intensity thresholding and connected component analysis) and subtracted from the scaled-SWI in order to exclude large vessels.For a prospective regional analysis of SWI-iron, a robust mean value was computed for every cortical and deep GM regions by fitting a Gaussian-distribution in the histogram of all voxels within the region. The statistical effect of our hypothesis (iron accumulation across diagnostic groups) was evaluated using a multivariate general linear model (GLM) with SWI-iron as dependent variable and group and age as covariates. No significant group differences were found for sex or years of education, and therefore these were not controlled for. Iron-rich brain regions appear hypointense (low signal intensity i.e. darker) on SWI images and therefore lower SWI intensity indicates higher iron concentration.


The differences in regional SWI-iron concentrations between HC, MCI and AD groups are presented in Fig. 1 and Fig. 2. Given that the cerebellum is not affected until the very late stage of AD, Fig. 1 confirmed no significant difference across HC, MCI and AD diagnostic groups. Compared to HC, significantly higher SWI-iron levels were observed in AD individuals (p<0.001) and MCI (p<0.05) in the middle Frontal lobe as well as combined Posterior Cingulate, Cuneus and Precuneus regions (Fig. 2).


In the present work, we estimated brain iron level in HC, MCI and AD subjects using SWI analysis. We observed a stepwise increment in SWI-iron across clinical categories, consistent with iron accumulation with disease progression. Furthermore, the higher level of SWI-iron in the middle Frontal lobe and Precuneus but the cerebellum of AD individuals may be associated with the higher presence of Aβ plaques. This study suggests that ventricle normalized SWI may be a potential imaging biomarker of Alzheimer’s disease. However, further validation with quantitative iron measures (i.e. QSM and GRE T2*) is required to fully understand the association between normalized SWI intensities and the underlying susceptibility effect.


No acknowledgement found.


1. Smith, M. A., Harris, P. L., Sayre, L. M, et al. Iron accumulation in Alzheimer disease is a source of redox-generated free radicals. Proceedings of the National Academy of Sciences 94, 18 (1997), 9866–9868.

2. Duce, J. A., Tsatsanis, A., Cater, M. A., et al. Iron-export ferroxidase activity of β-amyloid precursor protein is inhibited by zinc in Alzheimer’s disease. Cell 142, 6 (2010), 857–867.

3. Huang, X., Moir, R. D., Tanzi, R. E., et al. Redoxactive metals, oxidative stress, and Alzheimer’s disease pathology. Annals of the New York Academy of Sciences 1012, 1 (2004), 153–163.

4. Cheng, A.-L., Batool, S., Mccreary, C. R., et al.. Susceptibility-Weighted Imaging is More Reliable Than T2*-Weighted Gradient-Recalled Echo MRI for Detecting Microbleeds. Stroke 44, 10 (2013), 2782–2786.

5. Haller, S., Bartsch, A., Nguyen, D., et al. Cerebral microhemorrhage and iron deposition in mild cognitive impairment: susceptibility-weighted MR imaging assessment. Radiology 257, 3 (2010), 764–773.

6. Kirsch, W., Mcauley, G., Holshouser, B., et al. Serial susceptibility weighted MRI measures brain iron and microbleeds in dementia. Journal of Alzheimer’s Disease 17, 3 (2009), 599–609.

7. Tzourio-Mazoyer, N., Landeau, B., Papathanassiou, D., et al. Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. Neuroimage 15, 1 (2002), 273–289.

8. Tustison, N., Avants, B., Cook, P., et al. N4ITK: improved N3 bias correction. IEEE Transactions on Medical Imaging 29, 6 (2010), 1310–1320.


Table 1: Demographic information.

The amyloid status was based on a neocortical amyloid-β standardize uptake value ratio ≥ 1.5. Age information is mean ±standard deviation with (minimum age-maximum age).

Fig. 1: Figure above illustrates the proposed estimation of SWI-iron values across healthy control (HC), mild cognitive impairment (MCI) and Alzheimer’s disease individuals. The box plots present the median and adjacent quartiles, and the circles and dots represent subject data points. No significant change in SWI-iron was observed in the Cerebellum WM.

Fig. 2: Figures above illustrate regional SWI-iron mean differences among diagnostic groups. The values were adjusted for age and p-values were obtained by comparing each group to the HC. The significance level of each test are reported on the top as ‘*’ and ‘***’ reflecting p-values less than 0.05 and 0.001, respectively.

Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)