Towards robust – and accurate – QSM in cortical and sub-cortical regions of the human brain at 9.4T: the influence of masking
Gisela E Hagberg1,2, Elisa Tuzzi1, Joana Loureiro1, Thomas Ethofer1,3, Rolf Pohmann4, Jonas Bause4, Pascal Martin5, Marina Pavlova3, Marc Himmelbach6, Anja Zeller3, Christoph Laske3, Andreas J Fallgatter3, and Klaus Scheffler1,4

1Biomedical Magnetic Resonance, University Hospital Tübingen, Tübingen, Germany, 2High Field Magnetic Resonance, Max Planck Institute for Biological Cybernetics, Tübingen, Germany, 3Psychiatry and Psychotherapy, University Hospital Tübingen, Tübingen, Germany, 4Max Planck Institute for Biological Cybernetics, Tübingen, Germany, 5Neurology and Epileptology, Hertie Institute for Clinical Brain Research, Tübingen, Germany, 6Neuropsychology, Hertie Institute for Clinical Brain Research, Tübingen, Germany


Quantitative susceptibility mapping (QSM) targets a fundamental MR-parameter but is problematic due to the presence of a strong background and local field variations. These may cause multiple phase wraps which are particularly prominent at high fields and long echo-times. We propose automated tissue masking excluding brain areas with excessive phase wraps and show how this approach can improve the quality of QSM. Performance was evaluated with regard to iron quantification in subcortical and cortical areas, and was compared with R2* maps in the same 21 subjects aged 19-56y and literature values.


Quantitative susceptibility mapping (QSM) is hampered due to strong background field variations, therefore the product of the echo time and field is usually kept below 70ms·Tesla1. To achieve the high resolution at ultra-high fields, it may be difficult to remain within this limit. Here we investigate the quality of QSM in terms of accuracy and precision that could be achieved with multi-echo data at 9.4T, in comparison with literature data and R2* mapping of the same subjects to evaluate iron quantification.

Subjects and Methods

Subjects (19-56y N=21) volunteering to participate in the ERB approved study were scanned at 9.4T (Siemens Germany) with a 16ch transmit/31ch receive array2. The transmit field was mapped with AFI3 to acertain nominal flip angle in sub-cortical areas. Mono-polar multi-echo-GRE-images (TE=6:6:30ms; TR=35ms; nominal FA=11°; voxel=400x400x800μm; GRAPPA=2; PF=6/8; BW=240Hz) were used for QSM and MP2RAGE images (TI1/TI2=900/3500ms; FA=4/6°; TRGRE=6ms; TR=8894ms, 0.8mm isotropic voxels) for tissue segmentation and DARTEL–based adaptation (SPM12) of the Automatic Anatomical Labelling (AAL) and the Harvard-Oxford region-of-interest atlases in MNI space brought to native space images. Single-channel GRE-images were adaptively combined after phase offset correcting the third echo (TE=18ms) using the first two echo-images4. Laplacian phase unwrapping5 and background subtraction using RESHARP6, Tikhonov regularization of 10-3-10-12; or V-SHARP5 (SMV2-20 mm) were used. Three different masks (Fig1) based on the signal in magnitude images only (MM); or after additionally removing either all voxels with excessive phase wrapping (PB) or voxels in the region-of-interest ‘Rectus’ from AAL (noR) were generated. MM was obtained using BET in FSL, cut-off=0.1 followed by smoothing with a Gaussian filter with a FWHM 4 times the voxel-size. For PB, the absolute value of the Laplacian of the sign function was convolved with a sphere of size 5, and only voxels with values < 500 were retained. For noR the rectus AAL region was expanded and removed from the mask. The QSM maps were generated by the iterative least squares approach7. This pipeline was tested on the QSM challenge data set8 but including background correction which yielded RSME values of 80.5(Tk-12MM) 80.4(Tk-12PB), and 82.6(SMV12MM) and HFEN of 74.9, 75.0, and 74.8, respectively. QSM values in different Harvard-Oxford atlas regions were extracted from cortical voxels with a GM probability >98% while no such additional condition was used for the subcortical regions. Expected iron concentrations were calculated from Ref9.

Results and Discussion

The PB masks identified regions with excessive phase variations in the frontal region (Fig1). For QSM, the typical L-curve was observed with optimal settings at Tk-12 and SMV12. The traditional MM mask worked well in case of high regularization and small SMV filters, (Fig2 a-c), while lower regularization relied on adequate masking using PB or noR (Fig2 b,e,h). The lowest coefficient of variation (standard deviation divided by the average across subjects) was found using Tk-12PB in the Globus Pallidus (13%) while in the cortex this figure was at least more than twice as high (28%) obtained by SMV12PB in S1. QSM at the level of the basal ganglia was similar across all masks and for both background removal methods, although RESHARP retained more voxels on the brain surface than V-SHARP(Fig3), as expected for regions close to the border of the mask10. RESHARP was more robust than V-SHARP in this respect yielding a smaller discrepancy between the different masks (Table1), except for the motor cortex where the MM failed to yield plausible values in view of its relatively high iron content (Fig4). The relation between estimated iron and QSM values differed between RESHARP and V-SHARP and also slightly between masks. The highest regression coefficient was found for Tk-12PB, yielding 0.430 ppb per µg/kg iron, which also had the best goodness-of-fit performance. Nevertheless, this value is at the low end of previous observations11. We explain this result by our long echo time and high magnetic field, causing a diffusion averaging similar to what is usually observed for R2*. Indeed our change in R2* with increasing iron corresponds to previous studies11 as well as our iron-induced change in QSM.


We found a substantial improvement in accuracy and precision of QSM in high-field applications at long echo times through masking, targeting brain areas with excessive phase evolution.


Funding from EULAC-HEALTH T01-0118 is gratefully acknowledged


[1] Haacke EM et al., Magn reson Imag 33 1-25 (2015)

[2] Shajan G, Kozlov M, Hoffmann J, Turner R, Scheffler K, Pohmann R. A 16-channel dual-row transmit array in combination with a 31-element receive array for human brain imaging at 9.4 T Magn Reson Med. 2014;71(2):870-9.

[3] Yarnykh, V.L., 2007. Actual flip‐angle imaging in the pulsed steady state: a method for rapid three‐dimensional mapping of the transmitted radiofrequency field. Magnetic resonance in medicine 57, 192-200.

[4]Robinson SD, Bredies K, Khabipova D, Dymerska B, Marques JP, Schweser F. An illustrated comparison of processing methods for MR phase imaging and QSM: combining array coil signals and phase unwrapping. NMR Biomed. 2017;30(4).

[5] Li W, Wu B, Liu C. Quantitative susceptibility mapping of human brain reflects spatial variation in tissue composition, NeuroImage. 2011; 15; 55:1645

[6] Sun, H., Wilman, A.H., 2014. Background field removal using spherical mean value filtering and Tikhonov regularization. Magnetic resonance in medicine 71, 1151-1157

[7] Li W, Wu B, Liu C. Quantitative susceptibility mapping of human brain reflects spatial variation in tissue composition, NeuroImage. 2011; 15; 55:1645

[8] Langkammer C, Schweser F, Shmueli K, Kames C, Li X, Guo L, Milovic C, Kim J, Wei H, Bredies K, Buch S, Guo Y, Liu Z, Meineke J, Rauscher A, Marques JP, Bilgic B. Quantitative susceptibility mapping: Report from the 2016 reconstruction challenge.Magn Reson Med. 2018;79(3):1661-1673.

[19 Hallgren B and Sourander P. J Neurochem. 1958;3(1):41-51

[10] Schweser F, Robinson SD, de Rochefort L, Li W, Bredies K. An illustrated comparison of processing methods for phase MRI and QSM: removal of background field contributions from sources outside the region of interest. NMR Biomed. 2017;30(4).

[11] Duyn JH, Schenck J. Contributions to magnetic susceptibility of brain tissue. NMR Biomed. 2017;30(4):e3546


Figure 1: Sagittal magnitude (left), phase (middle) and smoothed Laplacian of the sign function of the phase (right). Green outline: Magnitude mask (MM); blue outline: after removal of voxels with excessive phase wraps (PB) and red outline: mask obtained from MM after removal of the anatomical region-of-interest ‘Rectus’ from the AAL atlas (noR).

Figure 2: Using the magnitude mask (a,b,c) yielded altered QSM value in the prefrontal cortex, less so with higher regularization (a: Tk-3) or by V-SHARP(c SMV12mm) than for less regularization (b Tk-12). With the PB mask that removed voxels with strong phase variation, (d,e,f) or with the alternative noR mask that removed all voxels pertaining to the AAL region ‘Rectus’ (g,h,i) these issues could be substantially improved

Figure 3: QSM at the level of the basal ganglia using the PB mask yielded similar results for both background removal methods a: RESHARP, Tk-12 and b: V-SHARP with SMV of 12mm, although the quantitative values differed significantly between thew two methods (Table1).

Figure 4: QSM values found in cortical and subcortical regions in 21 subjects (mean, standard error of means) for different regularization levels using RESHARP and V-SHARP with an SMV of 12mm. In most areas the restricted PB and noR masks lead to improved precision across subjects. In view of estimated iron concentrations, the values in sensory and motor cortex should be similar, which was the case for PB and noR but not for MM.

Table 1: Average and standard deviation across 21 subjects of the estimated iron concentration (from Ref 9), effective transverse relaxation time, R2* and the magnetic susceptibility, QSM using RESHARP Tk-12 or V-SHARP SMV12mm. The average regression coeffient±std for the dependence on estimated iron concentration is also given, together with the R2-statistic as a measure of the goodness-of-fit

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)