Investigation of intrascanner reproducibility of quantitative susceptibility mapping (QSM) and R2* at 3T
Xiang Feng1, Andreas Deistung1, Marianne Cleve1, Ferdinand Schweser2,3, and Juergen Reichenbach1

1Medical Physics Group, Institute of Diagnostic and Interventional Radiology, Jena University Hospital - Friedrich Schiller University Jena, Jena, Germany, 2Buffalo Neuroimaging Analysis Center, Department of Neurology, Jacobs School of Medicine and Biomedical Sciences, The State University of New York at Buffalo, Buffalo, NY, United States, 3MRI Molecular and Translational Research Center, Jacobs School of Medicine and Biomedical Sciences, The State University of New York at Buffalo, Buffalo, NY, United States


In the present work, we quatitatively assessed the reproducibility of Quantitative susceptibility mapping (QSM) and effective transverse relaxation rate (R2*) in healthy volunteers at 3T scanner during multiple scanning sessions. We also investigated the QSM normalization based on 2 reference structures, Cerebrospinal fluid (CSF) and frontal White Matter (fWM).Our results indicated that QSM using CSF as reference region reveals increased scan-rescan reliability than referencing with fWM.


Quantitative susceptibility mapping (QSM) [1-4] is a recently developed post-processing technique to quantify the bulk magnetic susceptibility of tissue using the phase images of complex-valued gradient echo (GRE) data, while the effective transverse relaxation rate, R2*, can be calculated from the magnitude images of multi-echo GRE data. QSM and R2* can be used to monitor iron content [5-6] and myelin [7] in brain tissue. However, repeated MRI scans of the same subject can result in different magnitude and phase images due to small variations in head orientation, imaging slab orientation, and scanner calibration. These small differences may lead to changes in estimates of susceptibility and R2* values of human brain structures. The purpose of this study was to assess the reproducibility of QSM and R2* mapping in healthy volunteers at 3T scanner during multiple scanning sessions.

Subjects and Methods

Acquisition: Eight young subjects (female, age 24.1±2.4 years) were scanned 4 times using the same measurement protocol on a 3T MRI scanner (Tim Trio, Siemens Healthcare, Erlangen, Germany). The time periods between two consecutive scans in a single 4-scan-series were variable and ranged from 3 (smallest) to 38 (largest) days. The acquisition protocol included whole brain T1-weighted imaging (MP-RAGE sequence) with isotropic spatial resolution of 1mm and whole brain multi-echo GRE imaging (TE1-TE6/ΔTE/TR=3.07ms-27.72ms/4.93ms/32ms, FA=20°, voxel size = 0.57mm×0.57mm×2.00mm).

Analysis: The multi-echo GRE phase images were processed with sophisticated harmonic artifact reduction for phase data (SHARP) [8] and homogeneity enabled incremental dipole inversion (HEIDI) [9] to calculate susceptibility maps. R2* maps were calculated by mono-exponential fitting of the signal decay of GRE magnitude images. Six subcortical structures, including putamen, globus pallidus, caudate nucleus, accumbens, hippocampus, and thalamus, were segmented using FIRST [10] incorporating a dedicated hybrid contrast and nonlinear registration (HC-nlFIRST framework) [11]. Cerebrospinal fluid (CSF) in the ventricles was segmented using FMRIB’s Automated Segmentation Tool (FAST) on T1-weighted images, and then transformed to the GRE image space. Additionally, the region of interest (ROI) for frontal WM (fWM) was manually defined in MNI space and then warped to the GRE image space for each subject using the affine transformation matrix and inverse warping field generated from the HC-nlFIRST framework.

ROI-based analysis of reproducibility: Referencing of susceptibility values was compared with respect to fWM and CSF, where the global offset for QSM was determined by the mean value of the reference region. Mean and standard deviations (SD) of susceptibility values and R2* of the subcortical structures determined from all repetitions were statistically compared using analysis of variances (ANOVA) and intra-class correlation (ICC).


Fig. 1 illustrates exemplary susceptibility and R2* maps obtained from the same subject during the 4 repeated scans, showing visual similarity in depicting the subcortical structures. Table 1 summarizes the mean value and inter-subject standard deviation of QSM, R2* and volumes for 6 segmented structures. Fig. 2 shows the subcortical segmentation results from the HC-nlFIRST method corresponded to the anatomical boundary. Subcortical segmentation revealed no significant volume differences for each segmented subcortical structure across scan sessions (p>0.85). Furthermore, no significant differences in mean values of susceptibility (p>0.05) and R2* (p>0.75) were observed across scan sessions. Fig. 3 demonstrates that histograms of fWM and CSF have similar distributions and mean susceptibility values of the globus pallidus across subjects for the different scanning dates show less variation if CSF is used as reference region compared to fWM. This finding is consistent for nearly all subcortical structures (except thalamus) indicated by higher p-values (suggesting more insignificant differences across measurements) and ICC (suggesting increased correlation across repeated measurements) values of susceptibility values referenced to CSF than the ones referenced to fWM (Table 2). In addition, reliability of R2* in subcortical structures over repeated scans is indicated by the high p-values and ICC vales shown in Table 2.


Excellent reproducibility of QSM and R2* mapping across multiple repeated scan sessions at a single 3T scanner was demonstrated qualitatively (visual inspection) and quantitatively (ANOVA and ICC values), suggesting their potential utility in longitudinal studies and clinical applications. Since QSM yields relative rather than absolute values of magnetic susceptibility due to the relative nature of GRE phase, susceptibility values have to be examined relative to a particular reference region or tissue. ANOVA and ICC of subcortical structures (except thalamus) indicate that QSM with CSF as reference region reveals increased inter-scan reliability than referencing with fWM. The lower p-values and ICC values of the susceptibility referenced to CSF in the thalamus may result from its heterogeneous representation on susceptibility maps across scans and subjects. Further work is required to solve this discrepancy.


No acknowledgement found.


[1] Reichenbach JR et al, 2015. Clin Neuroradiol. 22:225-30. [2] Wang Y et al, 2015. Magn. Reson. Med. 73:82-101. [3] Haacke EM et al, 2015. Magn Reson Imaging. 33:1-25. [4] Liu C et al., 2015. J Magn Reson Imaging. 42:23-41. [5] Langkammer C et al, 2012. Neuroimage. 62(3):1593-9. [6] Langkammer C et al, 2010. Radiology. 257(2):455-62. [7] Liu C et al., 2011.Neuroimage. 56(3):930-8. [8] Schweser F et al., 2011. NeuroImage. 54(4):2789-807. [9] Schweser F et al., 2012. NeuroImage. 62(3):2083-100. [10] Patenaude B et al., 2011. NeuroImage. 56(3):907-22. [11] Feng X et al., 2015. Proc. ISMRM. 2481.


Fig.1: Axial susceptibility and R2* maps obtained in 4 scan sessions from the same subject demonstrating good visual similarity in the depiction of subcortical structures. (a) – (d) Susceptibility maps, where ‘DX’ represents the scanning date. (e) – (f) Corresponding R2* maps.

Fig.2: Automatically segmented ROIs (b,d) of subcortical structures with HC-nlFIRST. Comparison of these segmentations with the corresponding susceptibility (a) and R2* map (c) reveals accurate alignment of the anatomical subcortical structure boundary (blue: globus pallidus, purple: putamen, yellow: hippocampus, green: thalamus, cyan-blue: caudate nucleus).

Fig.3: Quantitative analysis of normalization of QSM. (a-b) The histograms of susceptibility values in the CSF and fWM are shown from 2 random selected subjects in 4 scan series. Mean susceptibility values in globus pallidus (GP) referenced to (c) fWM and (d) CSF are presented as boxplots for all subjects.

Table 1: Summary of mean values, inter-subject standard deviation of susceptibility values, R2* and volumes in the 6 subcortical structures.

Table 2: Summary of p-value of the ANOVA and ICC values from 4 scan-series in the 6 subcortical structures.

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