Quantitative susceptibility mapping (QSM) and R2* in the human brain at 3 T

ZEMEDI-10710; No. of Pages 13

ARTICLE IN PRESS

ORIGINAL PAPER

Quantitative susceptibility mapping (QSM) and R2* in the human brain at 3 T Evaluation of intra-scanner repeatability Xiang Feng 1,∗ , Andreas Deistung 1,2,3 , Jürgen R. Reichenbach 1,4 1

Medical Physics Group, Institute of Diagnostic and Interventional Radiology, Jena University Hospital – Friedrich Schiller University Jena, Jena, Germany 2 Section of Experimental Neurology, Department of Neurology, Essen University Hospital, Essen, Germany 3 Erwin L. Hahn Institute for Magnetic Resonance Imaging, University Duisburg-Essen, Essen, Germany 4 Michael Stifel Center for Data-driven and Simulation Science Jena, Friedrich Schiller University Jena, Germany Received 22 December 2016; accepted 19 May 2017

Abstract Quantitative susceptibility mapping (QSM) and the effective transverse relaxation rate (R2 *) can be used to monitor iron and myelin content in brain tissue, which are both subject to changes in many neurological diseases but also during healthy aging. In this study, we quantitatively assessed the repeatability of QSM and R2 * by applying four independent scans in eight young healthy, female subjects on a 3 T MRI scanner. Since QSM does not yield absolute values for bulk magnetic susceptibilities, we additionally investigated the influence of the choice of a reference brain region for susceptibility by computing susceptibility differences with respect to five different brain structures (whole brain, frontal white matter (fWM), internal capsule (IC), cerebrospinal fluid (CSF) in the lateral ventricle, cortical gray matter (cGM)). The intra-class correlation coefficient (ICC), variance ratio (VR) and repeatability coefficient (RC) were used to evaluate the repeatability of the calculated susceptibility differences and the R2 * values in six different subcortical brain structures. Linear regression was used to analyze the correlation between susceptibility differences and R2 *. We found that the susceptibility differences with respect to each investigated reference region (0.868 ≤ mean ICC ≤ 0.914) and the R2 * values

Quantitative Suszeptibilitätskartierung (QSM) und R2 * im menschlichen Gehirn bei 3T Eine Intra-Scanner Reproduzierbarkeitsstudie Zusammenfassung Die Kartierung der magnetischen Suszeptibilität (quantitative susceptibility mapping, QSM) und der effektiven transversalen Relaxationsrate (R2 *) ermöglicht die nicht-invasive Charakterisierung des Eisen- und Myelingehalts in Hirngewebe, die bei verschiedenen neurologischen Erkrankungen, aber auch während des gesunden Alterns, Veränderungen unterliegen. In der vorliegenden Studie wurde die Wiederholbarkeit der Suszeptibilitäts- und R2 *-Kartierung durch wiederholtes Scannen (n = 4) von acht jungen, gesunden, weiblichen Probandinnen mit einem 3T-Magnetresonanztomographen quantitativ beurteilt. Da die Suszeptibilitätskartierung keine absoluten magnetischen Suszeptibilitätswerte liefert, wurde zusätzlich der Einfluss verschiedener anatomischer Referenzstrukturen (komplettes Gehirn, frontale weiße Substanz, capsula interna, zerebrospinale Flüssigkeit im

∗ Corresponding author: Xiang Feng, Medical Physics Group, Institute of Diagnostic and Interventional Radiology, Jena University Hospital – Friedrich Schiller University Jena, Philosophenweg 3, 07743 Jena, Germany. E-mail: [email protected] (X. Feng).

Z. Med. Phys. xxx (2017) xxx–xxx http://dx.doi.org/10.1016/j.zemedi.2017.05.003 www.elsevier.com/locate/zemedi

ARTICLE IN PRESS 2

X. Feng et al. / Z. Med. Phys. xxx (2017) xxx–xxx

(mean ICC = 0.923) were highly repeatable across the four times repeated scans. With consistently higher ICC, higher VR and lower RC, whole brain and cGM appeared to be the two most suitable reference regions for QSM with respect to repeatability.

lateralen Ventrikel, kortikale graue Substanz) auf die Ergebnisse der damit ermittelten Suszeptibilitätsdifferenzen untersucht. Die Wiederholbarkeit der berechneten Suszeptibilitätsdifferenzen und R2 *-Werte wurde in sechs verschiedenen subkortikalen Hirnstrukturen mit Hilfe des Intraklassen-Korrelationskoeffizienten, des Varianzverhältnisses und des Reproduzierbarkeitskoeffizienten bewertet. Lineare Regressionen und Korrelationen zwischen den unterschiedlich referenzierten Suszeptibilitätsdifferenzen und den R2 *-Werten wurden durchgeführt. Die Suszeptibilitätsdifferenzen waren über die vierfachen Messungen bezüglich jeder untersuchten anatomischen Referenzstruktur (0,868 ≤ Mittelwert ICC ≤ 0,914) und R2 * (Mittelwert ICC = 0,923) hoch reproduzierbar. Mit durchweg höheren Intraklassen-Koeffizienten, höheren Varianzverhältnissen und niedrigeren Reproduzierbarkeitskoeffizienten ist eine Referenzierung mit dem Mittelwert der Suszeptibilität über das Gesamthirn oder über die kortikale graue Substanz im Hinblick auf die Wiederholbarkeit der Suszeptibilitätskartierung am besten geeignet.

Keywords: Repeatability, Quantitative susceptibility mapping, Effective transverse relaxation, Magnetic resonance imaging

Schlüsselwörter: Wiederholbarkeit, quantitative Suszeptibilitätskartierung, effektive transversale Relaxation, Magnetresonanztomographie

Introduction Quantitative susceptibility mapping (QSM) [1–6] is a recently developed MRI post-processing technique to noninvasively quantify the bulk magnetic susceptibility of tissue by exploiting the phase of the MR signal from T2 *-weighted gradient echo (GRE) sequences. To date, QSM has already been widely applied to assess magnetic susceptibility sources in brain tissue, including iron [7], myelin [8] and calcifications [9–11], metabolic oxygen consumption [12,13] and taskrelated blood oxygenation level variations [14,15] in normal subjects as well as in patients with intracranial hemorrhages [16,17], various neuro-degenerative diseases, such as multiple sclerosis [18], Parkinson’s disease [19,20], Huntington’s disease [21] and Alzheimer’s disease [22]. Since multi-echo GRE pulse sequences are commonly employed for QSM data acquisition, the effective transverse relaxation rate, R2 *, can additionally be deduced from the magnitude data as a valuable by-product of [23]. R2 * represents an alternative measure to estimate the iron load in brain tissue and has been validated in both human and primate studies [24–26]. One significant problem of using QSM in cross-sectional and longitudinal studies is that the singularity at the origin of the k-space representation of the unit dipole response introduces an arbitrary, region-independent offset in the reconstructed susceptibility map [27]. Consequently, QSM is only capable to provide relative rather than absolute values of

magnetic susceptibility. This significant obstacle is typically overcome by referencing the reconstructed susceptibility values to a specific region of the brain. The topic of choosing a suitable reference region has already been tackled in one previous cross-sectional [28] and longitudinal study [29]. The authors suggested to use frontal white matter (fWM) [28], the posterior limb of the internal capsule (IC) or cerebrospinal fluid (CSF) in the ventricle area as referencing structure [29]. It has also been suggested to employ the reconstructed susceptibility values directly after field-to-source inversion as these values are intrinsically referenced to the mean susceptibility of the whole tissue region used for susceptibility computation (e.g., the whole brain tissue in case of brain exams) and no obvious systematic bias was observed between the analysis of the susceptibility as a function of age with and without referencing to CSF [30]. However, so far no studies have been conducted to evaluate thoroughly the effect of different references to the magnetic susceptibility in cross-sectional or longitudinal studies. Another important concern is that repeated independent MRI scans of the same subject will most likely result in different magnitude and phase images due to small variations of the subject’s head position, imaging slab orientation, and/or MRI scanner calibration. These small differences may, in turn, lead to changes in the estimates of susceptibility and R2 * values in the different human brain structures. Thus, evaluation of scan–rescan repeatability (i.e., the same imaging

ARTICLE IN PRESS X. Feng et al. / Z. Med. Phys. xxx (2017) xxx–xxx

measurement performed on the same subject using the same protocol with the same scanner at the same center) and reproducibility (i.e., imaging measurements performed on the same subjects using the same protocol with different scanners with, e.g., different field strengths, vendors, sites or reconstruction algorithms) [31] would be of utmost importance for the establishment of QSM and R2 * as quantitative imaging biomarkers in further applications. Several recent studies reported high to excellent reproducibility of QSM across different sites [32], field strengths (1.5 T and 3 T) [33,34], scanners [33] and reconstruction methods [35]. An intra-scanner scan–rescan test of QSM and R2 * mapping with five control subjects has already been performed in a previous study by exploiting voxel- and structure-wise differences among the repetitions [36]. However, a comprehensive repeatability study of both QSM and R2 * with multiple repetitions using the same scanner at the same site, which is a common study scenario in academic and clinical studies, has not been conducted so far. Consequently, the main goals of the present study were (i) to characterize the repeatability of both susceptibility and R2 * mapping of the brain in healthy volunteers at 3 T resulting from different scans at different days and (ii) to determine the influence of different reference regions on QSM repeatability.

Subjects and methods Subjects Eight young and healthy subjects without any history of neurological or psychiatric diseases (female; right-handed; mean age ± standard deviation, 24.1 ± 2.4 years; age range, 22–29 years) were recruited. The local ethics committee had approved the study and written informed consent was collected from all participating subjects. Acquisition Each subject was scanned four times on different days using the same measurement protocol on a 3 T whole-body MRI scanner (Tim Trio, Siemens Healthcare, Erlangen, Germany) with a 12-channel head coil. The time intervals between two consecutive scans varied for each subject and ranged from 3 to 38 days. The acquisition protocol included whole brain T1 -weighted imaging for automatic tissue segmentation and whole brain multi-echo GRE imaging for QSM and R2 * calculations. The former was performed using a magnetization-prepared rapid gradient echo (MP-RAGE) sequence with isotropic spatial resolution of 1 mm (repetition time (TR) = 2300 ms, echo time (TE) = 2.63 ms, inversion time (TI) = 1100 ms, flip angle (FA) = 7◦ , band width (BW) = 199 Hz/pixel, acceleration factor of two in parallel imaging, acquisition time of 5 min 21 s). The sequence parameters of the multi-echo GRE sequence included 6 echoes (monopolar readout), TE1 –TE6 /TE 3.07 ms–27.72 ms/4.93 ms, TR = 32 ms,

3

FA = 20◦ , BW = 507 Hz/pixel, voxel size = 0.57 mm × 0.57 mm × 2.00 mm, resulting in an acquisition time of 7 min 47 s. Data preprocessing Quantitative susceptibility maps were computed from the multi-echo GRE phase images. To this end, phase images were unwrapped for each echo, converted to frequency maps and combined across the different echo times according to [37]. Background frequency contributions were eliminated using sophisticated harmonic artifact removal for phase data with 10 different spherical kernels with varying radii ranging from 0.57 mm to 5.7 mm (V-SHARP [38], regularization using high-pass filter with cut-off frequency of 0.01 mm−1 [39]). Susceptibility maps were subsequently reconstructed using homogeneity enabled incremental dipole inversion (HEIDI) [40]. The maximum number of inner loop iterations and the minimum value of iterative changes of the NESTA solver for sparse recovery were set to 400 and 0.01, respectively. R2 * maps were calculated by mono-exponential fitting of the signal decay of the multi-echo GRE magnitude data using logarithmic calculus [41]. The GRE magnitude images (first echo) were aligned to the T1 -weighted images with rigid registration using SPM8 (Statistical Parametric Mapping, http://www.fil.ion.ucl.ac. uk/spm/software/spm8/). Six subcortical brain structures (putamen, globus pallidus, caudate nucleus, accumbens, hippocampus and thalamus) were automatically segmented using the HC-nlFIRST method [42]. Briefly, the HC-nlFIRST method creates a hybrid contrast by combining T1 -weighted images and susceptibility maps to improve the delineation of the subcortical brain structures (see Fig. 1), and uses nonlinear registration to align these images to the Montreal Neurological Institute (MNI) template. The established software tool FSL-FIRST [43] was applied to the warped hybrid contrast images and segmented the subcortical structures in MNI space. Subsequently, the segmented ROIs were transformed from MNI space to the native T1 -weighted image space by applying the inverse warping transform with nearest neighbor interpolation. The final step consisted in transforming the subcortical ROIs from the native T1 -weighted image space to the QSM/R2 * space by inverting the initial rigid registration. We decided to apply FIRST on the hybrid contrast because (i) it slightly improves subcortical segmentation compared to conventional T1 -weighted images [42], and (ii) subcortical segmentation using susceptibility maps [44] is far less established with no independent reproducibility analyses performed so far. Reference regions for QSM normalization We selected five different brain regions for referencing the susceptibility maps, including whole brain tissue, frontal white matter (fWM), posterior limb of internal capsule (IC),

ARTICLE IN PRESS 4

X. Feng et al. / Z. Med. Phys. xxx (2017) xxx–xxx

Figure 1. Axial view of susceptibility maps (referenced to the whole brain tissue, a) and T1-weighted images (b) of the same anatomical region and examination. Linearly combining the signal intensities of the magnetic susceptibility and T1-weighted image to a hybrid contrast image (c) improves subcortical delineation. The arrows indicate superior depiction of the globus pallidus in the hybrid contrast image compared to the T1-weighted image.

cerebrospinal fluid (CSF) and cortical gray matter (cGM). The volume of interest (VOI) of the whole brain was given by the brain tissue voxels that met the QSM reliability criterion associated with the phase pre-processing as introduced in [45]. Since we are not aware of any established software tools to segment automatically fWM or IC, VOIs for both fWM and IC were bilaterally defined once on three adjacent slices of the MNI template (by X.F.) and then warped to the QSM/R2 * space of the individual subjects. Furthermore, CSF and gray matter (containing both cortical and subcortical gray matter) were automatically segmented on T1 -weighted images using FMRIB’s Automated Segmentation Tool (FAST) [46]. The CSF and GM segmentation results were transformed to the QSM/R2 * space and further restricted to only include the lateral ventricle and cortical GM (i.e., excluding subcortical gray matter), respectively. Finally, using Freeview (Freesurfer, https://surfer.nmr.mgh.harvard.edu/fswiki/FreeviewGuide) all VOI outlines were superposed on the susceptibility and R2 * maps for visual inspection and manually refined using Freeview’s ‘voxel edit’ tool if necessary. To normalize the susceptibility maps, mean susceptibility values of the five VOIs were calculated and subtracted from the reconstructed susceptibility maps. As a result, we obtained five susceptibility maps (referenced to the whole brain, fWM, IC, CSF and cGM, respectively) and one R2 * map.

of the mean values for each of these modalities was estimated with the intra-class correlation coefficient (ICC) and the variance ratio (VR) according to [31,48]: ICC =

VR =

σb2 σw2

(1)

(2)

where σb2 is the between-subject variance and σw2 the withinsubject variance. As shown in Eq. (1), ICC provides a relative ratio of the between-subject variance to the total variance (σt2 = σb2 + σw2 ), and is intrinsically restricted to the range from 0 to 1. VR is the ratio of between-subject variance and within-subject variance and has thus no fixed range. The higher these two metrics, the more repeatable QSM and R2 * are among the different scans. The consistency of the subcortical segmentation was assessed by calculating both the ICC and VR for the volumes of the individual subcortical structures. We also computed the within-subject standard deviation (wSD) and the repeatability coefficient (RC) together with the corresponding 95% confidence interval (CI) according to [48]:

Analysis of repeatability To evaluate visually how similar QSM and R2 * maps were between the scans, we calculated the structural similarity index (SSIM) [47], which is based on the three characteristics of luminance, contrast and structural similarity between images, and enables an objective metric for assessing perceptual image quality. To investigate the repeatability of QSM and R2 * in the six subcortical structures, we calculated the mean values and standard deviations of these structures for each subject, scan and modality (five differently referenced susceptibility maps and one R2 * map). The scan–rescan repeatability

σb2 σb2 + σw2

wSD =

 σw2 = σw

(3)

and 

RC = 1.96

2σw2 = 2.77σw

(4)

wSD is the unbiased estimate of the (root of square) betweensubject variance σw2 , whereas the interpretation of RC is that the difference between any two repeated scans of the same subject is expected to range from −RC to +RC with a probability

ARTICLE IN PRESS X. Feng et al. / Z. Med. Phys. xxx (2017) xxx–xxx

5

Figure 2. Axial view of susceptibility (referenced to the whole brain tissue, a–d) and corresponding R2 * maps (h–k) obtained in four different scan sessions of the same subject over a period of 40 days. Susceptibility and R2 * difference maps between the 2nd , 3rd or 4th scan and the first scan are presented in (e–g) and (l–n), respectively. ‘DX’ in the top row indicates the scanning date. Excellent visual similarity in the depiction of subcortical structures is demonstrated. The susceptibility (referenced to other structures, not shown) revealed the same visual impression.

ARTICLE IN PRESS 6

X. Feng et al. / Z. Med. Phys. xxx (2017) xxx–xxx

of 95%. Thus, smaller RC values indicate improved repeatability among scans, which also holds true for wSD. Under the assumption that both magnetic susceptibility and R2 * in the subcortical structures are mainly determined by iron content, we calculated the correlation between mean magnetic susceptibility (referenced to the five different reference regions) and mean R2 * across the subcortical regions and the repetitions. To characterize the influence of the different reference regions on the variations of magnetic susceptibility, we investigated the distribution pattern when plotting magnetic susceptibility as a function of R2 *. Towards this end, Pearson’s correlation coefficient, r, regression slope and intercept were computed using linear weighted total least-squares regression, where the standard deviations of the susceptibilities and R2 * values extracted from the corresponding subcortical structures were incorporated as weights.

Results Fig. 2 illustrates exemplarily susceptibility and R2 * maps obtained from four repeated scans of the same subject demonstrating excellent visual similarity regarding depiction of the subcortical structures. This is also supported by the low absolute values in the difference images between repetitions (Fig. 2e–f, l–n) and by the structure similarity indices between the different repetitions with values larger than 0.83 and 0.72 for the susceptibility and R2 * maps, respectively (see Table 1). Table 2 summarizes the mean values and intersubject standard deviations of the susceptibility differences (with respect to the five different reference structures), R2 * and the volumes for each subcortical structure. The subcortical volumes exhibit rather small changes with a maximum and minimum relative change of 4.14% for the accumbens and 0.94% for the thalamus, respectively. Fig. 3(a–d) illustrates the reference VOIs used to calculate the susceptibility differences and the corresponding histograms of the susceptibility distribution. The histograms of the smaller regions (fWM and IC) are narrower compared to the ones of the larger regions (cGM, CSF and whole brain tissue). In addition, the subcortical segmentation result by applying the HC-nlFIRST method is also

presented in Fig. 3(e), demonstrating accurate correspondence with the underlying anatomy. The results of the ICC and VR analysis are summarized in Tables 3 and 4, respectively. Susceptibility differences referenced to the whole brain tissue (ICC = 0.914, VR = 19.799) and R2 * (ICC = 0.923, VR = 18.064) both show high ICC and VR values, indicating excellent repeatability. High ICCs and VRs were also obtained for the susceptibility differences referenced to cGM and CSF, while lower values resulted when selecting fWM and IC as reference regions. The ultra-high ICC (ICC = 0.988) and VR values (VR = 212.341) of the subcortical volumes underline the consistent and reliable subcortical segmentation with the HC-nlFIRST method. The RC values of each modality are plotted in Fig. 4. Due to the linear relationship between wSD and RC, wSD reveals exactly the same tendency as RC in Fig. 4 (not shown). Susceptibility differences referenced to the whole brain tissue and R2 * exhibit the smallest values. CSF- and cGM-referenced susceptibilities reveal slightly higher values, whereas fWMand IC-referenced susceptibilities show further increased values. Since all the metrics employed (ICC, VR, wSD, RC) rely on different kinds of variances, we summarized the corresponding standard deviations in detail in Table 5. Linear regression between susceptibility differences and R2 *, illustrated in Fig. 5, revealed correlation coefficients above 0.82 for the investigated six subcortical structures, underlining the dominant contribution of iron to both magnetic susceptibility and R2 *. The results of the regression analysis including Pearson’s regression coefficient, regression slope and intercept are listed in Fig. 5.

Discussion We investigated the repeatability of both magnetic susceptibility and R2 * mapping in selected subcortical structures across young healthy females. In line with similar recent studies [32–36], we were able to show that both magnetic susceptibility and R2 * are highly consistent across repeated imaging experiments. In contrast to the mentioned studies, which investigated magnetic susceptibility either with no explicit referencing or referencing to ventricular CSF, we

Table 1 Summary of structural similarity index (mean ± standard deviation [SD]) for susceptibility and R2 * between any two scans. The overall mean and SD of structural similarity index were also computed among all the pairs of two scans. Abbreviations: WB – whole brain tissue, fWM – frontal white matter, IC – internal capsule, CSF – cerebrospinal fluid, cGM – cortical gray matter.

QSM

R2 *

Reference regions

Scan 1–scan 2

Scan 1–scan 3

Scan 1–scan 4

Scan 2–scan 3

Scan 2–scan 4

Scan 3–scan 4

Overall mean ± SD

WB fWM IC CSF cGM

0.879 ± 0.013 0.897 ± 0.015 0.874 ± 0.052 0.891 ± 0.015 0.880 ± 0.013

0.879 ± 0.016 0.900 ± 0.014 0.878 ± 0.060 0.892 ± 0.018 0.880 ± 0.016

0.878 ± 0.026 0.899 ± 0.021 0.805 ± 0.183 0.891 ± 0.026 0.880 ± 0.025

0.885 ± 0.017 0.903 ± 0.019 0.806 ± 0.181 0.896 ± 0.019 0.886 ± 0.017

0.879 ± 0.014 0.901 ± 0.011 0.778 ± 0.196 0.891 ± 0.019 0.880 ± 0.014

0.878 ± 0.021 0.896 ± 0.024 0.824 ± 0.118 0.892 ± 0.023 0.880 ± 0.021

0.88 ± 0.02 0.90 ± 0.02 0.83 ± 0.14 0.89 ± 0.02 0.88 ± 0.02

0.717 ± 0.016

0.714 ± 0.028

0.718 ± 0.043

0.725 ± 0.029

0.721 ± 0.021

0.715 ± 0.033

0.72 ± 0.03

ARTICLE IN PRESS X. Feng et al. / Z. Med. Phys. xxx (2017) xxx–xxx

7

Table 2 Summary of the mean values and inter-subject standard deviations of susceptibility differences, R2 * and volume for each scan and the six investigated subcortical structures. Mean susceptibility values and standard deviations are listed for the five differently referenced susceptibility maps. Reference region Accumbens

QSM (ppb)

R2 * (s−1 ) Volume (mm3 ) Caudate

QSM (ppb)

R2 * (s−1 ) Volume (mm3 ) Globus pallidus

QSM (ppb)

R2 * (s−1 ) Volume (mm3 ) Hippocampus

QSM (ppb)

QSM (ppb)

R2 * (s−1 ) Volume (mm3 ) Thalamus

QSM (ppb)

R2 * (s−1 ) Volume (mm3 )

Scan 2

Scan 3

Scan 4

Whole brain fWM IC CSF cGM

22.86 39.42 41.38 12.54 20.95 15.48 992

± ± ± ± ± ± ±

10.95 10.61 14.75 10.96 10.77 2.22 203

25.17 43.23 44.31 15.25 23.04 14.83 1026

± ± ± ± ± ± ±

7.68 10.24 8.92 7.69 7.56 1.68 216

27.04 46.27 44.20 16.97 24.79 14.64 1017

± ± ± ± ± ± ±

8.31 11.09 14.63 8.37 7.99 1.23 174

23.75 40.77 39.58 13.66 21.76 14.62 1018

± ± ± ± ± ± ±

8.46 10.91 12.10 8.09 8.28 1.63 192

Whole brain fWM IC CSF cGM

39.69 56.25 58.20 29.37 37.78 20.28 7710

± ± ± ± ± ± ±

3.31 3.93 5.35 2.96 3.54 0.92 980

37.73 55.79 56.87 27.81 35.60 20.10 7672

± ± ± ± ± ± ±

3.97 3.05 8.56 3.63 4.24 1.13 998

37.70 56.92 54.85 27.63 35.45 20.17 7611

± ± ± ± ± ± ±

5.63 4.67 10.19 6.46 5.94 1.57 989

39.72 56.74 55.56 29.64 37.74 20.01 7675

± ± ± ± ± ± ±

4.56 5.37 9.34 6.87 4.74 1.45 1049

Whole brain fWM IC CSF cGM

114.51 131.08 133.03 104.19 112.60 33.52 3523

± ± ± ± ± ± ±

14.27 15.17 13.35 17.03 14.59 3.00 444

113.68 131.74 132.82 103.76 111.55 32.91 3597

± ± ± ± ± ± ±

16.69 17.82 13.76 18.83 16.79 2.71 414

115.48 134.70 132.63 105.40 113.22 33.02 3431

± ± ± ± ± ± ±

16.08 16.28 15.33 18.41 16.22 2.63 281

116.03 133.05 131.86 105.94 114.04 33.09 3531

± ± ± ± ± ± ±

15.74 17.61 14.86 18.96 15.97 2.69 349

Whole brain fWM IC CSF cGM

−1.98 14.58 16.53 −12.30 −3.90 17.35 8225

± ± ± ± ± ± ±

3.85 2.50 7.25 4.55 3.95 3.19 422

0.86 18.92 20.00 −9.06 −1.27 17.38 8177

± ± ± ± ± ± ±

4.54 6.08 9.54 3.70 4.72 3.11 518

−3.22 16.00 13.93 −13.29 −5.47 17.18 8266

± ± ± ± ± ± ±

6.42 6.51 13.49 5.95 6.69 2.27 471

−1.66 15.36 14.18 −11.74 −3.65 17.15 8180

± ± ± ± ± ± ±

6.42 5.69 13.62 7.21 6.53 2.91 429

Whole brain fWM IC CSF cGM

36.09 52.65 54.60 25.77 34.17 22.14 10273

± ± ± ± ± ± ±

7.49 5.73 8.82 8.75 7.88 1.19 1293

36.93 54.99 56.07 27.01 34.80 21.83 10235

± ± ± ± ± ± ±

8.01 7.72 9.62 8.50 8.54 1.32 1153

36.38 55.60 53.53 26.31 34.13 22.18 10164

± ± ± ± ± ± ±

8.53 7.16 14.10 8.38 9.12 1.42 1081

36.79 53.81 52.63 26.70 34.80 21.86 10175

± ± ± ± ± ± ±

7.61 5.65 14.59 9.24 7.84 1.30 1174

Whole brain fWM IC CSF cGM

3.38 19.94 21.89 −6.95 1.46 20.06 16310

± ± ± ± ± ± ±

3.92 4.56 6.46 4.46 3.86 106 957

2.02 20.08 21.16 −7.90 −0.11 20.45 16377

± ± ± ± ± ± ±

2.41 4.52 7.35 2.53 2.66 1.26 1052

2.96 22.18 20.11 −7.11 0.71 20.39 16244

± ± ± ± ± ± ±

3.08 4.45 9.35 3.53 3.38 2.20 1079

3.68 20.70 19.52 −6.40 1.69 20.31 16295

± ± ± ± ± ± ±

1.76 5.81 9.42 4.27 2.07 1.36 1042

R2 * (Hz) Volume (mm3 ) Putamen

Scan 1

additionally considered the influence of the reference region on susceptibility and found whole-brain or cGM referencing to be superior. We exclusively investigated young healthy female volunteers to eliminate age, sex and disease related variances. Since it has been demonstrated that cross-site errors are not significantly larger than within-site errors [32] and susceptibility measures are consistent across different magnetic field strengths [33,34], we decided to conduct our study with one

and the same MRI system. Operator dependent bias in determining the subcortical structures was efficiently reduced by applying automated segmentation as indicated by ICC values of the subcortical structure volumes larger than 0.96 (Table 3). Unlike the previous studies that applied MEDI and iLSQR for QSM reconstruction [32–35], we applied HEIDI to solve the ill-posed inverse field-to-susceptibility problem. The mean ICC of 0.914 (referenced to the whole-brain, Table 3) is in good agreement with the median ICC of 0.946 determined

ARTICLE IN PRESS 8

X. Feng et al. / Z. Med. Phys. xxx (2017) xxx–xxx

Figure 3. Different reference structures for QSM normalization. Top row: (a) frontal white matter, (b) posterior limb of internal capsule, (c) cerebrospinal fluid, (d) cortical gray matter. Automatically segmented subcortical ROIs (blue: globus pallidus, purple: putamen, yellow: hippocampus, green: thalamus, cyan-blue: caudate nucleus, orange: accumbens) are shown in (e). The bottom row shows the histograms of the susceptibility values in the reference structures of frontal white matter (fWM), internal capsule (IC), cerebrospinal fluid (CSF), cortical gray matter (cGM) and whole brain tissue. Table 3 Summary of intra-class correlation coefficients (ICC) of the different region referenced susceptibilities, R2 * values and volumes for the subcortical structures. Mean and standard deviation (SD) of intra-class correlation coefficients across the subcortical structures are displayed in the bottom row. Susceptibility differences with respect to the whole brain, frontal white matter (fWM), internal capsule (IC), cerebrospinal fluid (CSF) and cortical gray matter (cGM) are presented, respectively. χ Reference structure

R2 *

Volume

Whole brain

fWM

IC

CSF

cGM

Accumbens Caudate Globus pallidus Hippocampus Putamen Thalamus

0.920 0.820 0.979 0.899 0.973 0.892

0.881 0.822 0.974 0.860 0.952 0.935

0.917 0.759 0.973 0.824 0.908 0.825

0.919 0.841 0.982 0.878 0.970 0.878

0.920 0.831 0.979 0.892 0.976 0.889

0.855 0.921 0.971 0.968 0.954 0.866

0.988 0.998 0.967 0.981 0.997 0.997

Mean ± SD

0.914 ± 0.059

0.904 ± 0.059

0.868 ± 0.078

0.911 ± 0.056

0.914 ± 0.057

0.923 ± 0.051

0.988 ± 0.012

Table 4 Summary of the variance ratio (VR) of the susceptibilities differences (χ) computed using different reference structures, R2 * and volumes for the subcortical structures. Mean and standard deviation (SD) of the VR across the subcortical structures are given in the bottom row. Susceptibility differences with respect to the whole brain, frontal white matter (fWM), internal capsule (IC), cerebrospinal fluid (CSF) and cortical gray matter (cGM) are presented, respectively. χ Reference structure

R2 *

Volume

Whole brain

fWM

IC

CSF

cGM

Accumbens Caudate Globus pallidus Hippocampus Putamen Thalamus

11.565 4.938 44.146 8.716 42.048 7.378

7.404 4.623 37.688 6.138 20.012 14.296

11.119 3.157 36.587 4.681 9.895 4.706

11.391 5.287 53.831 7.176 32.290 7.169

11.461 4.901 46.544 8.227 40.776 8.005

5.904 11.626 33.634 30.014 20.748 6.458

80.745 523.736 28.891 52.672 303.292 283.707

Mean ± SD

19.799 ± 18.185

15.027 ± 12.525

11.691 ± 12.603

19.524 ± 19.565

19.986 ± 18.545

18.064 ± 11.970

212.341 ± 193.609

ARTICLE IN PRESS X. Feng et al. / Z. Med. Phys. xxx (2017) xxx–xxx

9

Table 5 Summary of total standard deviation (SD), within-subject SD and between-subject SD for the subcortical structures across four repetitions for susceptibility values referenced to different anatomical structures and R2 *. Abbreviations: fWM – frontal white matter, IC – internal capsule, CSF – cerebrospinal fluid, cGM – cortical gray matter, ppb – parts per billion. R2 * (unit: s−1 )

QSM (unit: ppb) Reference structure Whole brain

fWM

IC

CSF

cGM

16.1263 7.2890 30.3685 9.7277 15.2560 5.0245

18.6678 6.8818 32.2417 9.3809 12.4874 8.9645

22.9467 12.8344 27.4866 18.3219 21.1718 13.1255

16.0164 8.6063 35.5986 9.6800 16.6401 6.4110

15.7194 7.7099 30.7928 9.9453 16.0922 5.3798

2.884 2.299 5.293 5.491 2.453 2.550

4.5494 2.9912 4.5198 3.1208 2.3252 1.7359

6.4397 2.9023 5.1835 3.5113 2.7242 2.2921

6.5917 6.2949 4.4834 7.6872 6.4142 5.4948

4.5501 3.4323 4.8075 3.3853 2.8840 2.2431

4.4530 3.1739 4.4658 3.2740 2.4897 1.7928

1.098 0.647 0.899 0.986 0.526 0.934

15.4713 6.6470 30.0302 9.2135 15.0778 4.7151

17.5220 6.2399 31.8223 8.6990 12.1866 8.6665

21.9795 11.1847 27.1185 16.6313 20.1768 11.9200

15.3565 7.8922 35.2725 9.0687 16.3883 6.0058

15.0755 7.0262 30.4673 9.3910 15.8984 5.0723

2.667 2.206 5.216 5.402 2.396 2.373

SDa

Total Accumbens Caudate Globus pallidus Hippocampus Putamen Thalamus Within-subject SDb Accumbens Caudate Globus pallidus Hippocampus Putamen Thalamus Between-subject SDc Accumbens Caudate Globus pallidus Hippocampus Putamen Thalamus a b c

Total SD equals to square root of sum of within-subject variance and between-subject variance. Within-subject SD corresponds to σw . Between-subject SD corresponds to σ b .

across iron-rich GM nuclei (including caudate, putamen, GP, red nucleus, substantia nigra, subthalamic nucleus) on susceptibility maps reconstructed with SHARP (radius = 6 voxels) followed by MEDI among four repeated scans of 14 healthy

Figure 4. Repeatability coefficient (RC) analysis of susceptibility differences (unit: ppb) and R2 * (unit: s-1 ) in subcortical structures. The error bars indicate the 95% confidence intervals of RC. Abbreviations: Puta – putamen, Caud – caudate, Accu – accumbens, Hipp – hippocampus, GP – globus pallidus, Thal – thalamus, WB – whole brain tissue, fWM – frontal white matter, IC – internal capsule, CSF – cerebrospinal fluid, cGM – cortical gray matter.

young subjects [35]. Our slightly lower mean ICC may be due to the fact that we included brain structures with substantially lower iron content, namely thalamus and hippocampus, in our analysis, while Santin et al. [35] focused only on structures with high iron content. One further explanation might be the different applied field-to-susceptibility inversion algorithms. Since MEDI incorporates a priori information to the full susceptibility k-space whereas HEIDI does so only to a small sub-region of the susceptibility k-space, maps reconstructed with MEDI are typically more smoothed [40]. Direct comparison to the other reproducibility studies mentioned before [32–34,36] is difficult due to the different study design. Of special note, however, is that Lin et al. demonstrated an average imprecision level of approximately 5 to 10 ppb when inspecting MEDI susceptibility maps [32]. We obtained substantially different ICC values for the different susceptibility reference regions. Mean ICC was lowest and highest when referencing to IC (mean ICC = 0.868) and whole brain tissue or cGM (mean ICC = 0.914), respectively. In accordance to [49,50], the high ICCs for the subcortical brain structures with values larger than 0.9 (no matter which reference structure is used) indicate excellent repeatability for the iron-rich nuclei (globus pallidus and putamen), while structures with relatively lower iron content

ARTICLE IN PRESS 10

X. Feng et al. / Z. Med. Phys. xxx (2017) xxx–xxx

Figure 5. Regression analysis between susceptibility differences and R2 * in subcortical structures for all subjects and repetitions. Susceptibility differences, χ, between (a) whole brain tissue, (b) frontal white matter, (c) internal capsule, (d) CSF, and (e) cortical GM are plotted as function of R2 *.

(accumbens, caudate, hippocampus and thalamus) can still be ascribed good repeatability. Referencing to fWM or IC the ICCs ranged between 0.759 and 0.974, indicating good to excellent repeatability. The results of VR (Table 4) and RC (Fig. 4) also support usage of whole-brain, CSF or cGM as reference region rather than IC or fWM. The susceptibility histograms of the reference regions (Fig. 3) furthermore suggest that the reference VOI size is critical in providing high ICCs in an equally aged cohort, i.e., larger regions result in higher ICCs. However, tissue composition of a reference region needs to be considered individually, particularly when comparing subjects of different ages and/or diseases. The two investigated white matter reference regions exhibit substantial different structural architecture. fWM is composed of crossing fibers containing thinner myelin sheaths and smaller axon diameters as well as larger extracellular spaces. It interconnects neocortical areas and connects to other brain areas further downstream [51]. The posterior limb of the IC, on the other hand, is characterized by tightly packed and highly myelinated parallel running fiber bundles of the cortical-spinal tract with larger axon diameters [51,52]. Consequently, susceptibilities measured in IC are more influenced by susceptibility anisotropy of the myelin sheaths [53] and structural anisotropy [54,55]. Both WM regions are furthermore likely to be affected by demyelination with subjects of different ages and/or neurological and psychiatric disorders. The relatively small volume of these WM reference regions

(Fig. 3) makes them also more sensitive to inaccuracies in automatic segmentation. While we measured inferior repeatability (substantially lower ICC and VR, substantially higher RC) of susceptibility differences with respect to these two WM regions, other studies suggested both regions as excellent choices to reference magnetic susceptibility. Straub et al. [29] determined the IC to be one of the two most suited reference regions when analyzing follow-up measurements of melanoma patients. A possible explanation for this contradictory finding may be that the authors investigated patients with a broad age range (21–87 years), who additionally underwent therapy between the repeated measurements. Deistung et al. [28] analyzed the mean of the standard deviations across 168 gray matter regions across similarly aged healthy subjects for different reference regions (e.g. CSF, putamen, large veins, global WM, frontal deep WM, occipital WM) as an indicator for inter-subject variation, which was found to be lowest with fWM as reference. In contrast to the present study, the susceptibility maps were computed based on the multiangle orientation COSMOS approach [56]. Since COSMOS susceptibility maps are expected to have averaged out contributions due to susceptibility sources other than those with isotropic susceptibility [40], they are less prone to anisotropic sources compared to single angle susceptibility computational approaches. In line with findings by Straub et al. [29], we found good repeatability of susceptibility differences with respect to CSF.

ARTICLE IN PRESS X. Feng et al. / Z. Med. Phys. xxx (2017) xxx–xxx

Although ventricular CSF susceptibility seems to be age independent, it may, however, be influenced in neurodegenerative diseases, such as, e.g., multiple sclerosis, where changes in its composition may occur due to the presence of specific group of proteins or breakdown products of myelin. It should be also noted that CSF is not homogeneous as the ventricles are traversed by blood vessels and the iron-containing choroid plexuses [57]. In addition, pulsatile CSF flow may occur leading to flow-related phase shifts which in turn decrease the reliability of the susceptibility values. Last but not least, the size of the ventricles is age-dependent [58] and may substantially differ across subjects and patients, making automated delineation of the CSF in the ventricles in elderly subjects or patients difficult. As already stated referencing susceptibility to whole brain or cGM revealed consistently the highest ICCs and VRs as well as the lowest RCs across all subcortical structures. Whole brain referencing represents an intrinsic referencing when employing SHARP in combination with iterative inverse field-to-susceptibility solution strategies [30]. Since the susceptibility maps are referenced to the mean susceptibility of the whole brain, no further preparation of any specific VOI is required. Whole brain referencing is thus expected to work well in healthy, similarly aged subjects (as investigated in our study). However, in patients with abnormal iron load, severe hemorrhagic or calcified lesions or neurodegeneration the susceptibility distribution may globally shift, resulting in a disease related bias that may decrease the fidelity of QSM, especially in longitudinal patient studies. We propose cGM to be used as reference structure because the large volume provides robust estimation of the mean susceptibility. Furthermore, compared to deep GM nuclei [59] and white matter [60], cGM has substantially lower iron levels and myelin content, respectively. The less myelination of the cortex (compared to white matter) and the axonal arrangement with axons running both parallel and perpendicular to the cortex surface makes the susceptibility of the cortex less anisotropic [61]. Cortical gray matter, in contrast to subcortical regions, also exhibits almost no age-related changes of iron content [62]. Since iron is the dominating contributor to the susceptibility of the cerebral cortex [63,64], the mean susceptibility value across the cortex is barely affected by age. In addition, cortical GM is presumably less affected by neurological and psychiatric diseases than white matter or deep GM. Given all these beneficial properties to serve as reference region, one should, however, be aware that care has to be exercised regarding background field removal in the QSM reconstruction pipeline, because insufficient removal can result in non-susceptibility related variations and, thus, affect the reliability of susceptibility values. With recently introduced techniques [65,66], the size of the deconvolution kernel applied during background field removal is reduced when approaching brain boundaries to minimize convolution artifacts at the edges of the brain where the phase support ends.

11

We found excellent repeatability of R2 * maps with ICCs and VRs comparable to those of whole brain tissue or cortical gray matter referenced susceptibility. ICCs for R2 * are substantially higher to the ones reported [35] (ICC = 0.721). A possible reason for our high ICCs is most likely related to the different multi-echo gradient-echo acquisition protocol. While we used 6 echoes with a bandwidth of 507 Hz/px, the authors applied 12 echoes with a bandwidth of 1000 Hz/px, resulting in substantial decreased signal-to-noise ratios in the individual echoes and, thus, reduced fitting quality. As expected and confirmed already in previous in vivo [28,67] and in situ [67] studies, we observed a linear relationship between the magnetic susceptibility and R2 * for the iron containing subcortical structures that was more or less independent of the reference region (Fig. 5). In line with ICC, VR, and RC results, whole brain and cGM susceptibility referencing exhibited the highest correlation coefficients indicating the presence of least intra- and inter-subject susceptibility variations. There are several limitations to our study, which are related to the small number of participating subjects and the fact that we did not include elderly subjects or patients with different neurological diseases. Consequently, our findings may not be directly transferable to clinical studies and the choice of the most suited region for referencing QSM should be carefully considered in such cases. We only adopted one particular background field removal method and QSM reconstruction algorithm although a growing number of alternative algorithms are meanwhile available. This certainly needs to be investigated in future studies. Nevertheless, the highly reproducible results obtained here underline the promising potential of susceptibility and transverse relaxation rates to be used as biomarkers in both the healthy and diseased brain states.

Conclusion Excellent repeatability of susceptibility differences with respect to the whole brain, CSF and cortical gray matter as well as R2 * values has been demonstrated in a group of equallyaged volunteers across multiple scan sessions with a 3 T MRI scanner suggesting the use of these quantitative measures in cross-sectional and longitudinal studies. We propose to reference magnetic susceptibility with respect to cortical gray matter as the region is relatively large and less influenced by iron and myelin variations compared to other parts of the brain.

Acknowledgements The study was supported by the German Research Foundation (DE 2516/1-1), by a stipend (Landesgraduiertenstipendium) from the Graduate Academy of the Friedrich Schiller University Jena awarded to X.F. It was also supported by the bilateral PPP-USA 2015/2016 program (project 57131995 – Quantitative MRI for Neurological Studies) of the German Academic Exchange Service (DAAD) awarded to

ARTICLE IN PRESS 12

X. Feng et al. / Z. Med. Phys. xxx (2017) xxx–xxx

J.R.R. The authors thank Marianne Cleve for supporting this study regarding volunteer recruitment and data acquisition. [20]

References [1] Deistung A, Schweser F, Reichenbach JR. Overview of quantitative susceptibility mapping. NMR Biomed 2017;30:e3569. [2] Schweser F, Deistung A, Reichenbach JR. Foundations of MRI phase imaging and processing for Quantitative Susceptibility Mapping (QSM). Z Med Phys 2016;26(1):6–34. [3] Haacke EM, Liu S, Buch S, Zheng W, Wu D, Ye Y. Quantitative susceptibility mapping: current status and future directions. Magn Reson Imaging 2015;33(1):1–25. [4] Wang Y, Liu T. Quantitative susceptibility mapping (QSM): decoding MRI data for a tissue magnetic biomarker. Magn Reson Med 2015;73(1):82–101. [5] Reichenbach J, Schweser F, Serres B, Deistung A. Quantitative susceptibility mapping: concepts and applications. Clin Neuroradiol 2015;25(2):225–30. [6] Liu C, Li W, Tong KA, Yeom KW, Kuzminski S. Susceptibility-weighted imaging and quantitative susceptibility mapping in the brain. J Magn Reson Imaging 2015;42(1):23–41. [7] Langkammer C, Schweser F, Krebs N, Deistung A, Goessler W, Scheurer E, et al. Quantitative susceptibility mapping (QSM) as a means to measure brain iron? A post mortem validation study. NeuroImage 2012;62(3):1593–9. [8] Liu C, Li W, Johnson GA, Wu B. High-field (9.4 T) MRI of brain dysmyelination by quantitative mapping of magnetic susceptibility. NeuroImage 2011;56(3):930–8. [9] Chen W, Zhu W, Kovanlikaya I, Kovanlikaya A, Liu T, Wang S, et al. Intracranial calcifications and hemorrhages: characterization with quantitative susceptibility mapping. Radiology 2014;270(2):496–505. [10] Deistung A, Schweser F, Wiestler B, Abello M, Roethke M, Sahm F, et al. Quantitative susceptibility mapping differentiates between blood depositions and calcifications in patients with glioblastoma. PLoS One 2013;8(3):e57924. [11] Schweser F, Deistung A, Lehr BW, Reichenbach JR. Differentiation between diamagnetic and paramagnetic cerebral lesions based on magnetic susceptibility mapping. Med Phys 2010;37(10):5165–78. [12] Fan AP, Evans KC, Stout JN, Rosen BR, Adalsteinsson E. Regional quantification of cerebral venous oxygenation from MRI susceptibility during hypercapnia. Neuroimage 2015;104:146–55. [13] Zhang J, Liu T, Gupta A, Spincemaille P, Nguyen TD, Wang Y. Quantitative mapping of cerebral metabolic rate of oxygen (CMRO2) using quantitative susceptibility mapping (QSM). Magn Reson Med 2015;74(4):945–52. [14] Sun H, Seres P, Wilman AH. Structural and functional quantitative susceptibility mapping from standard fMRI studies. NMR Biomed 2017;30:e3619, http://dx.doi.org/10.1002/nbm.3619. [15] Balla DZ, Sanchez-Panchuelo RM, Wharton SJ, Hagberg GE, Scheffler K, Francis ST, et al. Functional quantitative susceptibility mapping (fQSM). NeuroImage 2014;100:112–24. [16] Sun HF, Kate M, Gioia LC, Emery DJ, Butcher K, Wilman AH. Quantitative susceptibility mapping using a superposed dipole inversion method: application to intracranial hemorrhage. Magn Reson Med 2016;76(3):781–91. [17] Chang SX, Zhang JW, Liu T, Tsiouris AJ, Shou J, Nguyen T, et al. Quantitative susceptibility mapping of intracerebral hemorrhages at various stages. J Magn Reson Imaging 2016;44(2):420–5. [18] Langkammer C, Liu T, Khalil M, Enzinger C, Jehna M, Fuchs S, et al. Quantitative susceptibility mapping in multiple sclerosis. Radiology 2013;267(2):551–9. [19] Barbosa JH, Santos AC, Tumas V, Liu M, Zheng W, Haacke EM, et al. Quantifying brain iron deposition in patients with Parkinson’s disease

[21]

[22]

[23]

[24]

[25]

[26]

[27]

[28]

[29]

[30]

[31]

[32]

[33]

[34]

[35]

[36]

[37]

using quantitative susceptibility mapping, R2 and R2*. Magn Reson Imaging 2015. Langkammer C, Pirpamer L, Seiler S, Deistung A, Schweser F, Franthal S, et al. Quantitative susceptibility mapping in Parkinson’s disease. PLoS One 2016;11(9):e0162460. Ng AC, Poudel G, Stout JC, Churchyard A, Chua P, Egan GF, et al. Iron accumulation in the basal ganglia in Huntington’s disease: cross-sectional data from the IMAGE-HD study. J Neurol Neurosurg Psychiatry 2016;87(5):545–9. Acosta-Cabronero J, Williams GB, Cardenas-Blanco A, Arnold RJ, Lupson V, Nestor PJ. In vivo quantitative susceptibility mapping (QSM) in Alzheimer’s disease. PLoS One 2013;8(11):e81093. Deistung A, Schafer A, Schweser F, Biedermann U, Gullmar D, Trampel R, et al. High-resolution mr imaging of the human brainstem in vivo at 7 Tesla. Front Hum Neurosci 2013;7:710. Ordidge RJ, Gorell JM, Deniau JC, Knight RA, Helpern JA. Assessment of relative brain iron concentrations using T2-weighted and T2*-weighted MRI at 3 Tesla. Magn Reson Med 1994;32(3): 335–41. Hardy PA, Gash D, Yokel R, Andersen A, Ai Y, Zhang Z. Correlation of R2 with total iron concentration in the brains of rhesus monkeys. J Magn Reson Imaging 2005;21(2):118–27. Langkammer C, Krebs N, Goessler W, Scheurer E, Ebner F, Yen K, et al. Quantitative MR imaging of brain iron: a postmortem validation study. Radiology 2010;257(2):455–62. Cheng YC, Neelavalli J, Haacke EM. Limitations of calculating field distributions and magnetic susceptibilities in MRI using a Fourier based method. Phys Med Biol 2009;54(5):1169–89. Deistung A, Schafer A, Schweser F, Biedermann U, Turner R, Reichenbach JR. Toward in vivo histology: a comparison of quantitative susceptibility mapping (QSM) with magnitude-, phase-, and R2*-imaging at ultra-high magnetic field strength. Neuroimage 2013;65:299–314. Straub S, Schneider TM, Emmerich J, Freitag MT, Ziener CH, Schlemmer HP, et al. Suitable reference tissues for quantitative susceptibility mapping of the brain. Magn Reson Med 2016, http://dx.doi.org/10.1002/mrm.26369 [Epub ahead of print]. Li W, Wu B, Batrachenko A, Bancroft-Wu V, Morey RA, Shashi V, et al. Differential developmental trajectories of magnetic susceptibility in human brain gray and white matter over the lifespan. Hum Brain Mapp 2014;35(6):2698–713. Raunig DL, McShane LM, Pennello G, Gatsonis C, Carson PL, Voyvodic JT, et al. Quantitative imaging biomarkers: a review of statistical methods for technical performance assessment. Stat Methods Med Res 2015;24(1):27–67. Lin PY, Chao TC, Wu ML. Quantitative susceptibility mapping of human brain at 3 T: a multisite reproducibility study. AJNR Am J Neuroradiol 2015;36(3):467–74. Hinoda T, Fushimi Y, Okada T, Fujimoto K, Liu C, Yamamoto A, et al. Quantitative susceptibility mapping at 3 T and 1.5 T: evaluation of consistency and reproducibility. Investig Radiol 2015;50(8):522–30. Deh K, Nguyen TD, Eskreis-Winkler S, Prince MR, Spincemaille P, Gauthier S, et al. Reproducibility of quantitative susceptibility mapping in the brain at two field strengths from two vendors. J Magn Reson Imaging 2015;42(6):1592–600. Santin M, Didier M, Valabrègue R, Yahia Cherif L, García-Lorenzo D, Loureiro de Sousa P, et al. Reproducibility of R2* and quantitative susceptibility mapping (QSM) reconstruction methods in the basal ganglia of healthy subjects. NMR Biomed 2016. Cobzas D, Sun HF, Walsh AJ, Lebel RM, Blevins G, Wilman AH. Subcortical gray matter segmentation and voxel-based analysis using transverse relaxation and quantitative susceptibility mapping with application to multiple sclerosis. J Magn Reson Imaging 2015;42(6):1601–10. Wu B, Li W, Avram AV, Gho SM, Liu C. Fast and tissue-optimized mapping of magnetic susceptibility and T2* with multi-echo and multishot spirals. Neuroimage 2012;59(1):297–305.

ARTICLE IN PRESS X. Feng et al. / Z. Med. Phys. xxx (2017) xxx–xxx

[38] Li W, Wu B, Liu C. Quantitative susceptibility mapping of human brain reflects spatial variation in tissue composition. NeuroImage 2011;55(4):1645–56. [39] Ozbay PS, Deistung A, Feng X, Nanz D, Reichenbach JR, Schweser F. A comprehensive numerical analysis of background phase correction with V-SHARP. NMR Biomed 2017;30(4):e3550. [40] Schweser F, Sommer K, Deistung A, Reichenbach JR. Quantitative susceptibility mapping for investigating subtle susceptibility variations in the human brain. NeuroImage 2012;62(3):2083–100. [41] Miller AJ, Joseph PM. The use of power images to perform quantitative analysis on low SNR MR images. Magn Reson Imaging 1993;11(7):1051–6. [42] Feng X, Deistung A, Dwyer MG, Hagemeier J, Polak P, Lebenberg J, et al. An improved FSL-FIRST pipeline for subcortical gray matter segmentation to study abnormal brain anatomy using quantitative susceptibility mapping (QSM). Magn Reson Imaging 2017;39:110–22. [43] Patenaude B, Smith SM, Kennedy DN, Jenkinson M. A Bayesian model of shape and appearance for subcortical brain segmentation. Neuroimage 2011;56(3):907–22. [44] Lim IA, Faria AV, Li X, Hsu JT, Airan RD, Mori S, et al. Human brain atlas for automated region of interest selection in quantitative susceptibility mapping: application to determine iron content in deep gray matter structures. Neuroimage 2013;82:449–69. [45] Schweser F, Deistung A, Lehr BW, Reichenbach JR. Quantitative imaging of intrinsic magnetic tissue properties using MRI signal phase: an approach to in vivo brain iron metabolism? NeuroImage 2011;54(4):2789–807. [46] Zhang YY, Brady M, Smith S. Segmentation of brain MR images through a hidden Markov random field model and the expectationmaximization algorithm. IEEE Trans Med Imaging 2001;20(1):45–57. [47] Wang Z, Bovik AC, Sheikh HR, Simoncelli EP. Image quality assessment: from error visibility to structural similarity. IEEE Trans Image Process 2004;13(4):600–12. [48] Barnhart HX, Barboriak DP. Applications of the repeatability of quantitative imaging biomarkers: a review of statistical analysis of repeat data sets. Transl Oncol 2009;2(4):231–5. [49] Cicchetti DV. Guidelines, criteria, and rules of thumb for evaluating normed and standardized assessment instruments in psychology. Psychol Assess 1994;6(4):284. [50] Portney LG, Watkins MP. Foundations of clinical research: applications to practice. FA Davis; 2015. [51] Groeschel S, Hagberg GE, Schultz T, Balla DZ, Klose U, Hauser TK, et al. Assessing white matter microstructure in brain regions with different myelin architecture using MRI. PLoS One 2016;11(11):e0167274. [52] Yagishita A, Nakano I, Oda M, Hirano A. Location of the corticospinal tract in the internal capsule at MR imaging. Radiology 1994;191(2):455–60.

13

[53] Li W, Wu B, Avram AV, Liu C. Magnetic susceptibility anisotropy of human brain in vivo and its molecular underpinnings. Neuroimage 2012;59(3):2088–97. [54] Wharton S, Bowtell R. Effects of white matter microstructure on phase and susceptibility maps. Magn Reson Med 2015;73(3):1258–69. [55] He X, Yablonskiy DA. Biophysical mechanisms of phase contrast in gradient echo MRI. Proc Natl Acad Sci U S A 2009;106(32): 13558–63. [56] Liu T, Spincemaille P, de Rochefort L, Kressler B, Wang Y. Calculation of susceptibility through multiple orientation sampling (COSMOS): a method for conditioning the inverse problem from measured magnetic field map to susceptibility source image in MRI. Magn Reson Med 2009;61(1):196–204. [57] Morris CM, Candy JM, Oakley AE, Bloxham CA, Edwardson JA. Histochemical distribution of non-haem iron in the human brain. Acta Anat (Basel) 1992;144(3):235–57. [58] Barron SA, Jacobs L, Kinkel WR. Changes in size of normal lateral ventricles during aging determined by computerized tomography. Neurology 1976;26(11):1011–3. [59] Hallgren B, Sourander P. The effect of age on the non-haemin iron in the human brain. J Neurochem 1958;3(1):41–51. [60] Guleria S, Kelly TG. Myelin, myelination, and corresponding magnetic resonance imaging changes. Radiol Clin N Am 2014;52(2):227–39. [61] Marques JP, Khabipova D, Gruetter R. Studying cyto and myeloarchitecture of the human cortex at ultra-high field with quantitative imaging: R 1, R 2 and susceptibility. NeuroImage 2016. [62] Ramos P, Santos A, Pinto NR, Mendes R, Magalhães T, Almeida A. Iron levels in the human brain: a post-mortem study of anatomical region differences and age-related changes. J Trace Elem Med Biol 2014;28(1):13–7. [63] Fukunaga M, Li TQ, van Gelderen P, de Zwart JA, Shmueli K, Yao B, et al. Layer-specific variation of iron content in cerebral cortex as a source of MRI contrast. Proc Natl Acad Sci U S A 2010;107(8): 3834–9. [64] Deistung A, Schäfer A, Schweser F, Reichenbach J. Cortical mapping of magnetic susceptibility and R2* reveals insights into tissue composition. In: Proceedings of the 23rd annual meeting ISMRM. 2015. p. 2015. [65] Topfer R, Schweser F, Deistung A, Reichenbach JR, Wilman AH. SHARP edges: recovering cortical phase contrast through harmonic extension. Magn Reson Med 2015;73(2):851–6. [66] Li W, Avram AV, Wu B, Xiao X, Liu C. Integrated Laplacian-based phase unwrapping and background phase removal for quantitative susceptibility mapping. NMR Biomed 2014;27(2):219–27. [67] Sun H, Walsh AJ, Lebel RM, Blevins G, Catz I, Lu JQ, et al. Validation of quantitative susceptibility mapping with Perls’ iron staining for subcortical gray matter. Neuroimage 2015;105:486–92.

Available online at www.sciencedirect.com

ScienceDirect