- Email: [email protected]
Changes of the apparent diffusion coefficient in brain diffusion-weighted images due to subject positioning: A simulation study
Journal of Neuroradiology (2015) 42, 150—155
Available online at
ScienceDirect www.sciencedirect.com
ORIGINAL ARTICLE
Changes of the apparent diffusion coefficient in brain diffusion-weighted images due to subject positioning: A simulation study S. Kumazawa ∗, D. Ushijima , H. Yabuuchi Department of health sciences, faculty of medical sciences, Kyushu university, 3-1-1, Maidashi, Higashi-ku, Fukuoka 812-8582, Japan Available online 1st April 2015
KEYWORDS Apparent diffusion coefficient; Echo planar imaging; Magnetic field inhomogeneity; Geometric distortion; Subject positioning
Summary Background and purpose: The purpose of this work was to investigate whether the cortical apparent diffusion coefficient (ADC) values derived from brain diffusion-weighted images vary with changes in the position of the subject against a static magnetic field. Materials and methods: To focus on the variations in ADC due to the change of subject positioning, a simulation was performed using a digital brain phantom. The magnetic field inhomogeneities in the digital phantom were calculated for each subject position while changing the angle between the direction of the static field and the head of the digital phantom. The angle was changed from 0 to 40 degrees at 10-degree intervals. For each angle, the diffusionweighted images were simulated based on magnetic resonance physics in which the magnetic field inhomogeneity was taken into account. The relative differences of average ADC values between the tilt angles were calculated to evaluate the variations in ADC. The Wilcoxon ranksum test was used for comparisons of ADC values between the tilt angles for each cortical region. Results: In the cortical regions distorted by magnetic field inhomogeneities, the average ADC values differed significantly according to the position of the subject (P < 0.05). The range of the relative differences in average ADC values in relation to the differences in subject positioning was approximately 1% to 12%. Conclusion: Our results suggest that subject positioning against a static field is one of the factors affecting the accuracy of cortical ADC measurements derived from brain diffusion-weighted images. © 2015 Elsevier Masson SAS. All rights reserved.
Abbreviations: ADC, apparent diffusion coefficient; DW, diffusion-weighted; MRI, magnetic resonance imaging; EPI, echo-planar imaging; CSF, cerebrospinal fluid; OM, orbito-meatal; WM, white matter; GM, gray matter. ∗ Corresponding author. Tel.: +81 92 642 6696; fax: +81 92 642 6696. E-mail address: s [email protected] (S. Kumazawa). http://dx.doi.org/10.1016/j.neurad.2015.01.003 0150-9861/© 2015 Elsevier Masson SAS. All rights reserved.
Changes of ADC due to subject positioning
Introduction Diffusion-weighted (DW) magnetic resonance imaging (MRI) provides information regarding the microstructure of tissues [1,2]. Because the apparent diffusion coefficient (ADC) calculated from DW images reflects the information on tissue microscopic features within a voxel, the ADC value has been used to differentiate between benign and malignant lesions [3—5], and to assess the treatment response [6—9]. In the brain, the cortical ADC values have been investigated in relation to both neurological and neurodegenerative diseases [10—12]. While the ADC has been widely used for these quantitative assessments, various pitfalls in ADC measurement have also been reported, such as eddy currents, subject motion, and geometric distortion in echo-planar imaging (EPI) [13,14]. In the case of ADC measurements in the cerebral cortex, the signal contamination by cerebrospinal fluid (CSF) due to a partial volume effect causes overestimation, because the ADC values in the CSF are approximately three times larger than those in the cortex [14,15]. Moreover, the geometric distortion may complicate the CSF partial volume effect in DW images [15—17]. EPI is commonly used for DW MRI, since it is not sensitive to subject motion during imaging [14]. However, it is well known that EPI images suffer from geometric distortion due to magnetic field inhomogeneity, and the pixel displacement caused by this distortion is proportional to the magnetic field inhomogeneity [18]. The magnetic field inhomogeneity, in turn, depends on the magnetic susceptibility distribution of the object placed in the magnetic field, with differences in the magnetic susceptibility between tissues, and particularly between tissue and air, causing variations in the induced magnetic field [19]. In the brain, the magnetic field inhomogeneity due to the magnetic susceptibility can be observed in the regions around air/tissue interfaces such as the frontal lobe base and paranasal sinuses. Because the static field perturbation is expressed by the magnetic susceptibility distribution of the object and the magnetic field, the change of the angle between the air/tissue interface and the static field leads to a change in the magnetic susceptibility-induced inhomogeneity. Truong et al. [20] demonstrated that tilting the head backwards significantly reduces the magnetic susceptibility-induced inhomogeneity in the region superior to the planum sphenoidale. This means that the magnetic field inhomogeneity is changed by the subject positioning relative to the static field. Previous studies [20,21] have reported on the relationship between the magnetic susceptibility-induced inhomogeneity and subject positioning. As described above, the magnetic field inhomogeneity causes a distortion in the EPI image, resulting in signal contamination by CSF and consequent overestimation of the ADC in the cortex. The purpose of our study was to investigate whether the cortical ADC values change along with changes in the subject positioning against the static field. It is known that the ADC values depend on many factors, such as image noise, imaging parameters, artifacts and MRI system hardware [13,14,22,23]. In this study, in order to focus on the variations in ADC due to the change of subject
151 positioning, we investigated the relationship between them using a simulator.
Materials and methods The EPI-based DW images were simulated using the 3D digital brain phantom introduced by the McConnell Brain Imaging Centre, Montreal Neurological Institute, McGill University [24]. The digital brain phantom consists of ten tissue types, and each voxel is assigned to each tissue type. The original volume data was converted into 128 × 128 × 128 volume data to be parallel to the orbitomeatal (OM) plane, and the resultant isotropic voxel was 2 mm3 . The magnetic susceptibility-induced inhomogeneities were calculated based on the digital brain phantom placed in the virtual static field using the susceptibility voxel convolution method [19]. The EPI-based DW images were simulated based on an MR signal equation in which the magnetic field inhomogeneity was taken into account.
Calculation of the field inhomogeneities When the object is placed in an MRI scanner, the differences in the magnetic susceptibility values between different tissues in the object cause variations in the induced magnetic field. In order to obtain the magnetic field distributions for a digital brain phantom placed in a virtual scanner, the susceptibility voxel convolution method [19] was used in this study. The magnetic susceptibility values described in the literature [19] were used as the values in each tissue type of the brain digital phantom. As shown in Fig. 1, the magnetic field distributions in the digital phantom were calculated by changing the angle between the direction of the static field and the OM plane in the digital phantom. A tilt angle of zero degrees meant that the OM plane was perpendicular to the static field (Fig. 1a). The angle was changed from 0 to 40 degrees at 10-degree intervals. Fig. 1k—o shows the magnetic field distributions at the slice level over the frontal sinus corresponding to the dotted lines in each sagittal view (Fig. 1f—j), respectively.
Simulation of the EPI-based DW images We simulated DW images according to the k-space trajectory in a single-shot spin echo EPI sequence. Let m and n represent the mth and nth sampling points in the readout (x) and the phase-encoding (y) directions in k-space, respectively (—M/2 ≤ m < M/2,—N/2 ≤ n < N/2). Let km and kn represent the spatial frequencies at the mth and the nth sampling points in k-space, respectively. Let Gx and Gb be the gradient in the x direction and the average blip gradient in the y direction during the duration , respectively. The MR signal S from a slice at location z0 is given by an integration of the signals from protons over the imaging volume
152
S. Kumazawa et al. A
S (km , kn ) =
(x, y, z) · ˇ (z − z0 ) exp −bD (x, y, z) exp −
· exp −j2km
x+
B (x, y, z) Gx
· exp −j2kn
where (x,y,z) is the proton density, D (x,y,z) is the diffusion coefficient in the tissue, T2 (x,y,z) is the transverse relaxation time, and B (x,y,z) is the magnetic field inhomogeneity at location (x,y,z). Here, ˇ (z) is the profile in the slice selection direction, A is a constant, b is the bfactor, and TE is the echo time. tx and ty are the time intervals between adjacent points in the readout and phaseencoding directions, respectively, Gy = Gb /ty. According to the discretization model in the literature [19], we make the assumption that the object is represented by a set of voxels, each of which is a homogeneous tissue. The DW images were simulated in a static field of 3 T using a digital brain phantom with tilt angles as shown in Fig. 1a—e. The magnetic field distributions calculated from the phantom with each tilt angle were used as B in equation (1). The simulated imaging parameters were as follows: TE = 70 ms, matrix size 128 × 128, field of view 256 mm, and 2.0 mm slice thickness. The pixel bandwidths in the readout and phase-encoding directions were 1445.3 Hz and 20.8 Hz, respectively. The and T2 values in
y+
TE + mtx + nty T2 (x, y, z) B (x, y, z) Gy
(1)
dxdydz,
each tissue type of the brain digital phantom were used as the values described in the literature [19]. The diffusion coefficients in white matter (WM), gray matter (GM), and CSF were 0.79, 0.87, and 3.34 [×10−3 mm2 /s], respectively [25]. The b-factors of 0 and 1000 were used, and ADC maps were generated.
Comparison of cortical ADC values We investigated whether the cortical ADC values were changed by tilting the head against the static field. In the digital brain phantom, neuroanatomical labels were assigned to 17 cortical regions on the right cerebral hemisphere that were painted manually by a radiologist (Fig. 2). In the presence of the magnetic field inhomogeneity B, these cortical regions are depicted in locations different from their original locations in the DW images. As can be
Figure 1 The magnetic field distributions in the digital brain phantom. a—e The different geometries of the phantom with respect to the static magnetic field. f—j The magnetic field distributions calculated from the phantom with each tilt angle. k—o The magnetic field distributions at the slice level over the frontal sinus corresponding to the dotted lines in each sagittal view, and the profiles along each center line.
Changes of ADC due to subject positioning
153
Figure 2 Seventeen cortical regions on the right cerebral hemisphere in the digital brain phantom. 1. Cingulate gyrus. 2. Superior frontal gyrus. 3. Middle frontal gyrus. 4. Inferior frontal gyrus. 5. Precentral gyrus. 6. Insular cortex. 7. Postcentral gyrus. 8. Transvers temporal gyrus. 9. Superior temporal gyrus. 10. Middle temporal gyrus. 11. Occipital gyri. 12. Visual cortex. 13. Cuneus. 14. Lingual gyrus. 15. Parahippocampal gyrus. 16. Caudate nucleus. 17. Thalamus.
seen in equation (1), the pixel displacements x and y at location (x,y,z) along the x and y directions are given by x =
B (x, y, z) B (x, y, z) ty and y = , Gx Gb
(2)
respectively. In order to measure the ADC values in these cortical regions in the distorted ADC maps due to the field inhomogeneity, the pixels in the neuroanatomical labelledimage were shifted based on the corresponding magnetic field distribution B and equation (2). For each cortical region, the average and standard deviation of ADC values were calculated. The average ADC values in each cortical region were compared between the images obtained from different tilt angles of the head in the digital phantom. We also calculated the relative differences of average ADC values between the tilt angles, which was given by (highest value—lowest value)/mean value [23].
Statistical analysis Statistical analysis was performed with the R Statistical Package (version 3.1.0, 2014; R Foundation for Statistical Computing, Vienna, Austria). The Wilcoxon rank-sum test was used for comparisons of ADC values between two tilt angles of all combinations for each cortical region. Statistical significance was defined as a P value of less than 0.05.
Results Fig. 3a—e shows the ADC maps at the slice level corresponding to the magnetic field distributions shown in Fig. 1k—o, respectively. The inhomogeneity-induced distortions in ADC maps were different from each other in the region close to the frontal sinus. Fig. 3f—j shows 17 cortical regions on the right cerebral hemisphere in each distorted image. The
pixels in each image were shifted using B values shown in Fig. 1k—o. Table 1 shows the averages and standard deviations of the ADC values of 17 cortical regions, among which the visual cortex (#12) had the highest ADC of 1.686 × 10−3 mm2 /s, and the inferior frontal gyrus (#4) had the lowest ADC of 0.972 × 10−3 mm2 /s. The average ADC over the 17 cortical regions was 1.129 × 10−3 mm2 /s. When the average ADC value for the superior frontal gyrus (#2) at 0 degrees was compared with those at 20, 30, and 40 degrees, the P values were 0.028, 0.032, and 0.031, respectively. The differences of the average ADC values between the tilt angles were statistically significant (P < 0.05). For the other regions, however, there was no significant difference (P > 0.05) between the tilt angles. The relative differences of average ADC values between the tilt angles ranged from 0.7% to 12.2%, and the average of the relative differences was 3.5%.
Discussion As shown in Fig. 3a—e, the distortions in ADC maps differed from each other in the region close to the frontal sinus. The corresponding regions in the magnetic field distributions were changed by tilting the head of the digital phantom backwards (Fig. 1k—o). The main contribution of the magnetic field inhomogeneity due to the magnetic susceptibility effect was the magnetic susceptibility distribution of the object along the axis parallel to the static magnetic field [19]. When a tissue was placed over the region of air along the axis parallel to the static magnetic field, the magnetic field inhomogeneity occurred at regions around the air/tissue interface. This effect was dependent on the positional relationship between them on the axis parallel to the static field. When viewed from the perspective of the positional relationship on the axis parallel to the static field, the tissues placed over the frontal sinus were moved to the region over the soft tissues by tilting the head backwards (arrows in Fig. 1f—j). Since the inhomogeneities at the frontal region were reduced by this displacement (profiles shown in Fig. 1k—o), the distortions in ADC maps were reduced. This decrease of the inhomogeneity in the frontal region induced by tilting the head backwards was in agreement with the simulation results in a previous study [20]. These results suggest that the differences of the subject positioning cause the changes of the degree of distortions in EPI-based DW images as well as the inhomogeneity in human data. In our DW MRI simulator, we assume that each voxel contains a single tissue type, i.e., we do not take into account the partial volume effect within a voxel. In this study, the diffusion coefficients in GM and CSF were set to 0.87 × 10−3 mm2 /s and 3.34 ×10−3 mm2 /s, respectively. Therefore, ideally, the ADC value in GM becomes 0.87 × 10−3 mm2 /s. As shown in Table 1, however, the averages of the ADC values of 17 cortical regions were higher than that, and increasing to between 12% and 94%. This was thought to be due to signal contamination by CSF in the process from MR signal generation to image reconstruction. In the large distortion region (#2: superior frontal gyrus), there were statistically significant differences in the average ADC values between the different subject positions (P < 0.05). The range of the
154
S. Kumazawa et al.
Figure 3 a—e ADC maps at the slice level corresponding to the magnetic field distributions shown in Fig. 1f—j, respectively, f—j 17 cortical regions on the right cerebral hemisphere in each distorted ADC map.
relative differences in average ADC values by the differences in the subject position was approximately 1% to 12%. In our analysis based on the simulation, these variations in ADC values were dependent on the differences of the subject positioning, and did not include other factors such as image noise and region of interest measurements by the operator. The relative difference of average ADC values by the difference of the subject positioning was at the same level as the relative difference of those by the difference of MRI hardware (4%—9%) [23]. Although the partial volume effect was not taken into account in our simulation, voxels in the data
Table 1
gathered from actual human subjects contain a mixture of multiple tissues due to the low resolution of the DW image [15—17]. Since the ADC value in CSF is approximately three times larger than that in GM, in the cortical region adjacent to CSF, the ADC values are increased by the partial volume effect between GM and CSF [14]. Therefore, the variations of ADC values by the difference of the subject positioning in the cortical region in the data from actual human subjects could be larger than those demonstrated by our simulation. Our results suggest that the positioning of the subject against the static field is an additional factor affecting the
Averages and standard deviations of ADC values of the 17 cortical regions.
Cortical region #
0 degree
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
1.113 1.157 1.069 0.974 1.196 1.130 1.278 1.001 0.947 1.025 1.240 1.584 1.034 1.088 1.035 1.044 1.129
± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
10 degree 0.357 0.330 0.230 0.189 0.310 0.335 0.389 0.206 0.158 0.289 0.424 0.428 0.249 0.269 0.250 0.268 0.329
1.079 1.242 1.056 0.974 1.201 1.123 1.274 1.012 0.951 1.029 1.212 1.684 1.022 1.091 1.043 1.051 1.129
± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
0.312 0.367 0.233 0.187 0.314 0.329 0.387 0.212 0.163 0.302 0.404 0.518 0.231 0.278 0.260 0.274 0.332
20 degree 1.078 1.310 1.063 0.974 1.230 1.125 1.259 0.994 0.952 1.039 1.249 1.686 1.048 1.112 1.047 1.061 1.145
± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
0.320 0.397* 0.253 0.186 0.325 0.316 0.356 0.199 0.174 0.308 0.463 0.565 0.271 0.289 0.260 0.274 0.350
30 degree 1.078 1.311 1.054 0.979 1.225 1.105 1.301 0.988 0.956 1.028 1.244 1.618 1.017 1.080 1.019 1.038 1.122
± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
0.317 0.374* 0.248 0.193 0.337 0.312 0.402 0.180 0.188 0.310 0.443 0.448 0.243 0.264 0.255 0.276 0.314
40 degree 1.065 1.312 1.035 0.972 1.201 1.123 1.288 1.008 0.944 1.036 1.250 1.672 1.023 1.081 1.005 1.026 1.168
± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
0.297 0.395* 0.203 0.182 0.307 0.337 0.377 0.213 0.162 0.309 0.455 0.576 0.237 0.268 0.254 0.260 0.367
Relative differences [%] 4.4 12.2 3.2 0.7 2.8 2.2 3.2 2.4 1.3 1.4 3.1 6.2 3.0 2.9 4.1 3.4 4.0
The cortical region number corresponds to the number referred to in Fig. 2. Data are represented as the average ADC [10−3 mm2 /s] ± standard deviation. * P < 0.05 compared with the results at 0 degrees.
Changes of ADC due to subject positioning accuracy of the cortical ADC measurements. In order to investigate the effects of this factor for clinical practice, we plan to analyze the ADC values in the actual human subject data. Moreover, the signal contamination by CSF can also affect the measurements in WM region [14]. In our future work, we plan to investigate the changes of ADC value in WM region due to the positioning of the subject against a static field and the partial volume effect between GM and WM. In this study, we investigated the relationship between the cortical ADC values and the subject positioning by tilting the head of a digital phantom against a static field. As shown in Fig. 1e, a tilt angle of 40 degrees corresponds to the position with the subject’s chin up, and is within the range of potential adult head movement in the head coil. In the case of infants and children, the range of movement in the head coil is wider than that in adults. Therefore, to reduce the variation of the cortical ADC values, it is important to pay attention to the position of the subject’s head against the static field, as well as the reproducibility of this positioning.
Conclusion We demonstrated that the cortical ADC values could vary substantially according to changes in the positioning of the subject against a static field. Our results suggest that the positioning of the subject against a static field is one of the factors affecting the accuracy of the cortical ADC measurements.
Disclosure of interest The authors declare that they have no conflicts of interest concerning this article.
Acknowledgements This work was supported by Japan Society for the Promotion of Science (JSPS) KAKENHI Grant Number 23500561.
References [1] Le Bihan D, Johansen-Berg H. Diffusion MRI at 25: exploring brain tissue structure and function. Neuroimage 2012;61:324—41. [2] Le Bihan D. Apparent diffusion coefficient and beyond: what diffusion MR imaging can tell us about tissue structure. Radiology 2013;268:318—22. [3] Guo Y, Cai YQ, Cai ZL, et al. Differentiation of clinically benign and malignant breast lesions using diffusion-weighted imaging. J Magn Reson Imaging 2002;16:172—8. [4] Woodhams R, Matsunaga K, Iwabuchi K, et al. Diffusionweighted imaging of malignant breast tumors: the usefulness of apparent diffusion coeffi cient (ADC) value and ADC map for the detection of malignant breast tumors and evaluation of cancer extension. J Comput Assist Tomogr 2005;29:644—9. [5] Liu L, Yin B, Geng DY, et al. Comparison of ADC values of intracranial hemangiopericytomas and angiomatous and anaplastic meningiomas. J Neuroradiol 2014;41:188—94.
155 [6] von Kummer R, Dzialowski I, Gerber J. Therapeutic efficacy of brain imaging in acute ischemic stroke patients. J Neuroradiol 2014 [in press]. [7] Song I, Kim CK, Park BK, et al. Assessment of response to radiotherapy for prostate cancer: value of diffusion-weighted MRI at 3 T. AJR Am J Roentgenol 2010;194:W477—82. [8] Foltz WD, Wu A, Chung P, et al. Changes in apparent diffusion coefficient and T2 relaxation during radiotherapy for prostate cancer. J Magn Reson Imaging 2013;37:909—16. [9] Park SH, Moon WK, Cho N, et al. Diffusion-weighted MR imaging: pretreatment prediction of response to neoadjuvant chemotherapy in patients with breast cancer. Radiology 2010;257:56—63. [10] Bozzao A, Floris R, Baviera ME, et al. Diffusion and perfusion MR imaging in cases of Alzheimer’s disease: correlations with cortical atrophy and lesion load. AJNR Am J Neuroradiol 2001;22:1030—6. [11] Ray KM, Wang H, Chu Y, et al. Mild cognitive impairment: apparent diffusion coefficient in regional gray matter and white matter structures. Radiology 2006;241:197—205. [12] Liu T, Young G, Huang L, et al. 76-space analysis of grey matter diffusivity: methods and applications. Neuroimage 2006;31:51—65. [13] Le Bihan D, Poupon C, Amadon A, et al. Artifacts and pitfalls in diffusion MRI. J Magn Reson Imaging 2006;24:478—88. [14] Jones DK, Cercignani M. Twenty-five pitfalls in the analysis of diffusion MRI data. NMR Biomed 2010;23:803—20. [15] Yoshiura T, Mihara F, Tanaka A, et al. Novel method to estimate and display cerebral cortical degeneration using diffusion-weighted magnetic resonance imaging. Magn Reson Med 2005;54:455—9. [16] Kumazawa S, Yoshiura T, Honda H, et al. Partial volume estimation and segmentation of brain tissue based on diffusion tensor MRI. Med Phys 2010;37:1482—90. [17] Kumazawa S, Yoshiura T, Honda H, et al. Improvement of partial volume segmentation for brain tissue on diffusion tensor images using multiple-tensor estimation. J Digit Imaging 2013;26:1131—40. [18] Jezzard P, Balaban RS. Correction for geometric distortion in echo planar images from B0 field variations. Magn Reson Med 1995;34:65—73. [19] Yoder DA, Zhao Y, Paschal CB, et al. MRI simulator with object-specific field map calculations. Magn Reson Imaging 2004;22:315—28. [20] Truong TK, Clymer BD, Chakeres DW, Schmalbrock P. Threedimensional numerical simulations of susceptibility-induced magnetic field inhomogeneities in the human head. Magn Reson Imaging 2002;20:759—70. [21] Heberlein K, Hu X. Improved shim by subject head positioning. In: Proceedings of the ninth annual meeting of the international Society of Magnetic Resonance in medicine. 2001. p. 1157. [22] Dietrich O, Heiland S, Sartor K. Noise correction for the exact determination of apparent diffusion coefficients at low SNR. Magn Reson Med 2001;45:448—53. [23] Sasaki M, Yamada K, Watanabe Y, et al. Variability in absolute apparent diffusion coefficient values across different platforms may be substantial: a multivendor, multi-institutional comparison study. Radiology 2008;249:624—30. [24] Cocosco CA, Kollokian V, Kwan RK, Evans AC. BrainWeb: online interface to a 3D MRI simulated brain database. NeuroImage 1997;5:S425 [part 2/4]. [25] Tanner SF, Ramenghi LA, Ridgway JP, et al. Quantitative comparison of intrabrain diffusion in adults and preterm and term neonates and infants. AJR Am J Roentgenol 2000;174:1643—9.
Available online at
ScienceDirect www.sciencedirect.com
ORIGINAL ARTICLE
Changes of the apparent diffusion coefficient in brain diffusion-weighted images due to subject positioning: A simulation study S. Kumazawa ∗, D. Ushijima , H. Yabuuchi Department of health sciences, faculty of medical sciences, Kyushu university, 3-1-1, Maidashi, Higashi-ku, Fukuoka 812-8582, Japan Available online 1st April 2015
KEYWORDS Apparent diffusion coefficient; Echo planar imaging; Magnetic field inhomogeneity; Geometric distortion; Subject positioning
Summary Background and purpose: The purpose of this work was to investigate whether the cortical apparent diffusion coefficient (ADC) values derived from brain diffusion-weighted images vary with changes in the position of the subject against a static magnetic field. Materials and methods: To focus on the variations in ADC due to the change of subject positioning, a simulation was performed using a digital brain phantom. The magnetic field inhomogeneities in the digital phantom were calculated for each subject position while changing the angle between the direction of the static field and the head of the digital phantom. The angle was changed from 0 to 40 degrees at 10-degree intervals. For each angle, the diffusionweighted images were simulated based on magnetic resonance physics in which the magnetic field inhomogeneity was taken into account. The relative differences of average ADC values between the tilt angles were calculated to evaluate the variations in ADC. The Wilcoxon ranksum test was used for comparisons of ADC values between the tilt angles for each cortical region. Results: In the cortical regions distorted by magnetic field inhomogeneities, the average ADC values differed significantly according to the position of the subject (P < 0.05). The range of the relative differences in average ADC values in relation to the differences in subject positioning was approximately 1% to 12%. Conclusion: Our results suggest that subject positioning against a static field is one of the factors affecting the accuracy of cortical ADC measurements derived from brain diffusion-weighted images. © 2015 Elsevier Masson SAS. All rights reserved.
Abbreviations: ADC, apparent diffusion coefficient; DW, diffusion-weighted; MRI, magnetic resonance imaging; EPI, echo-planar imaging; CSF, cerebrospinal fluid; OM, orbito-meatal; WM, white matter; GM, gray matter. ∗ Corresponding author. Tel.: +81 92 642 6696; fax: +81 92 642 6696. E-mail address: s [email protected] (S. Kumazawa). http://dx.doi.org/10.1016/j.neurad.2015.01.003 0150-9861/© 2015 Elsevier Masson SAS. All rights reserved.
Changes of ADC due to subject positioning
Introduction Diffusion-weighted (DW) magnetic resonance imaging (MRI) provides information regarding the microstructure of tissues [1,2]. Because the apparent diffusion coefficient (ADC) calculated from DW images reflects the information on tissue microscopic features within a voxel, the ADC value has been used to differentiate between benign and malignant lesions [3—5], and to assess the treatment response [6—9]. In the brain, the cortical ADC values have been investigated in relation to both neurological and neurodegenerative diseases [10—12]. While the ADC has been widely used for these quantitative assessments, various pitfalls in ADC measurement have also been reported, such as eddy currents, subject motion, and geometric distortion in echo-planar imaging (EPI) [13,14]. In the case of ADC measurements in the cerebral cortex, the signal contamination by cerebrospinal fluid (CSF) due to a partial volume effect causes overestimation, because the ADC values in the CSF are approximately three times larger than those in the cortex [14,15]. Moreover, the geometric distortion may complicate the CSF partial volume effect in DW images [15—17]. EPI is commonly used for DW MRI, since it is not sensitive to subject motion during imaging [14]. However, it is well known that EPI images suffer from geometric distortion due to magnetic field inhomogeneity, and the pixel displacement caused by this distortion is proportional to the magnetic field inhomogeneity [18]. The magnetic field inhomogeneity, in turn, depends on the magnetic susceptibility distribution of the object placed in the magnetic field, with differences in the magnetic susceptibility between tissues, and particularly between tissue and air, causing variations in the induced magnetic field [19]. In the brain, the magnetic field inhomogeneity due to the magnetic susceptibility can be observed in the regions around air/tissue interfaces such as the frontal lobe base and paranasal sinuses. Because the static field perturbation is expressed by the magnetic susceptibility distribution of the object and the magnetic field, the change of the angle between the air/tissue interface and the static field leads to a change in the magnetic susceptibility-induced inhomogeneity. Truong et al. [20] demonstrated that tilting the head backwards significantly reduces the magnetic susceptibility-induced inhomogeneity in the region superior to the planum sphenoidale. This means that the magnetic field inhomogeneity is changed by the subject positioning relative to the static field. Previous studies [20,21] have reported on the relationship between the magnetic susceptibility-induced inhomogeneity and subject positioning. As described above, the magnetic field inhomogeneity causes a distortion in the EPI image, resulting in signal contamination by CSF and consequent overestimation of the ADC in the cortex. The purpose of our study was to investigate whether the cortical ADC values change along with changes in the subject positioning against the static field. It is known that the ADC values depend on many factors, such as image noise, imaging parameters, artifacts and MRI system hardware [13,14,22,23]. In this study, in order to focus on the variations in ADC due to the change of subject
151 positioning, we investigated the relationship between them using a simulator.
Materials and methods The EPI-based DW images were simulated using the 3D digital brain phantom introduced by the McConnell Brain Imaging Centre, Montreal Neurological Institute, McGill University [24]. The digital brain phantom consists of ten tissue types, and each voxel is assigned to each tissue type. The original volume data was converted into 128 × 128 × 128 volume data to be parallel to the orbitomeatal (OM) plane, and the resultant isotropic voxel was 2 mm3 . The magnetic susceptibility-induced inhomogeneities were calculated based on the digital brain phantom placed in the virtual static field using the susceptibility voxel convolution method [19]. The EPI-based DW images were simulated based on an MR signal equation in which the magnetic field inhomogeneity was taken into account.
Calculation of the field inhomogeneities When the object is placed in an MRI scanner, the differences in the magnetic susceptibility values between different tissues in the object cause variations in the induced magnetic field. In order to obtain the magnetic field distributions for a digital brain phantom placed in a virtual scanner, the susceptibility voxel convolution method [19] was used in this study. The magnetic susceptibility values described in the literature [19] were used as the values in each tissue type of the brain digital phantom. As shown in Fig. 1, the magnetic field distributions in the digital phantom were calculated by changing the angle between the direction of the static field and the OM plane in the digital phantom. A tilt angle of zero degrees meant that the OM plane was perpendicular to the static field (Fig. 1a). The angle was changed from 0 to 40 degrees at 10-degree intervals. Fig. 1k—o shows the magnetic field distributions at the slice level over the frontal sinus corresponding to the dotted lines in each sagittal view (Fig. 1f—j), respectively.
Simulation of the EPI-based DW images We simulated DW images according to the k-space trajectory in a single-shot spin echo EPI sequence. Let m and n represent the mth and nth sampling points in the readout (x) and the phase-encoding (y) directions in k-space, respectively (—M/2 ≤ m < M/2,—N/2 ≤ n < N/2). Let km and kn represent the spatial frequencies at the mth and the nth sampling points in k-space, respectively. Let Gx and Gb be the gradient in the x direction and the average blip gradient in the y direction during the duration , respectively. The MR signal S from a slice at location z0 is given by an integration of the signals from protons over the imaging volume
152
S. Kumazawa et al. A
S (km , kn ) =
(x, y, z) · ˇ (z − z0 ) exp −bD (x, y, z) exp −
· exp −j2km
x+
B (x, y, z) Gx
· exp −j2kn
where (x,y,z) is the proton density, D (x,y,z) is the diffusion coefficient in the tissue, T2 (x,y,z) is the transverse relaxation time, and B (x,y,z) is the magnetic field inhomogeneity at location (x,y,z). Here, ˇ (z) is the profile in the slice selection direction, A is a constant, b is the bfactor, and TE is the echo time. tx and ty are the time intervals between adjacent points in the readout and phaseencoding directions, respectively, Gy = Gb /ty. According to the discretization model in the literature [19], we make the assumption that the object is represented by a set of voxels, each of which is a homogeneous tissue. The DW images were simulated in a static field of 3 T using a digital brain phantom with tilt angles as shown in Fig. 1a—e. The magnetic field distributions calculated from the phantom with each tilt angle were used as B in equation (1). The simulated imaging parameters were as follows: TE = 70 ms, matrix size 128 × 128, field of view 256 mm, and 2.0 mm slice thickness. The pixel bandwidths in the readout and phase-encoding directions were 1445.3 Hz and 20.8 Hz, respectively. The and T2 values in
y+
TE + mtx + nty T2 (x, y, z) B (x, y, z) Gy
(1)
dxdydz,
each tissue type of the brain digital phantom were used as the values described in the literature [19]. The diffusion coefficients in white matter (WM), gray matter (GM), and CSF were 0.79, 0.87, and 3.34 [×10−3 mm2 /s], respectively [25]. The b-factors of 0 and 1000 were used, and ADC maps were generated.
Comparison of cortical ADC values We investigated whether the cortical ADC values were changed by tilting the head against the static field. In the digital brain phantom, neuroanatomical labels were assigned to 17 cortical regions on the right cerebral hemisphere that were painted manually by a radiologist (Fig. 2). In the presence of the magnetic field inhomogeneity B, these cortical regions are depicted in locations different from their original locations in the DW images. As can be
Figure 1 The magnetic field distributions in the digital brain phantom. a—e The different geometries of the phantom with respect to the static magnetic field. f—j The magnetic field distributions calculated from the phantom with each tilt angle. k—o The magnetic field distributions at the slice level over the frontal sinus corresponding to the dotted lines in each sagittal view, and the profiles along each center line.
Changes of ADC due to subject positioning
153
Figure 2 Seventeen cortical regions on the right cerebral hemisphere in the digital brain phantom. 1. Cingulate gyrus. 2. Superior frontal gyrus. 3. Middle frontal gyrus. 4. Inferior frontal gyrus. 5. Precentral gyrus. 6. Insular cortex. 7. Postcentral gyrus. 8. Transvers temporal gyrus. 9. Superior temporal gyrus. 10. Middle temporal gyrus. 11. Occipital gyri. 12. Visual cortex. 13. Cuneus. 14. Lingual gyrus. 15. Parahippocampal gyrus. 16. Caudate nucleus. 17. Thalamus.
seen in equation (1), the pixel displacements x and y at location (x,y,z) along the x and y directions are given by x =
B (x, y, z) B (x, y, z) ty and y = , Gx Gb
(2)
respectively. In order to measure the ADC values in these cortical regions in the distorted ADC maps due to the field inhomogeneity, the pixels in the neuroanatomical labelledimage were shifted based on the corresponding magnetic field distribution B and equation (2). For each cortical region, the average and standard deviation of ADC values were calculated. The average ADC values in each cortical region were compared between the images obtained from different tilt angles of the head in the digital phantom. We also calculated the relative differences of average ADC values between the tilt angles, which was given by (highest value—lowest value)/mean value [23].
Statistical analysis Statistical analysis was performed with the R Statistical Package (version 3.1.0, 2014; R Foundation for Statistical Computing, Vienna, Austria). The Wilcoxon rank-sum test was used for comparisons of ADC values between two tilt angles of all combinations for each cortical region. Statistical significance was defined as a P value of less than 0.05.
Results Fig. 3a—e shows the ADC maps at the slice level corresponding to the magnetic field distributions shown in Fig. 1k—o, respectively. The inhomogeneity-induced distortions in ADC maps were different from each other in the region close to the frontal sinus. Fig. 3f—j shows 17 cortical regions on the right cerebral hemisphere in each distorted image. The
pixels in each image were shifted using B values shown in Fig. 1k—o. Table 1 shows the averages and standard deviations of the ADC values of 17 cortical regions, among which the visual cortex (#12) had the highest ADC of 1.686 × 10−3 mm2 /s, and the inferior frontal gyrus (#4) had the lowest ADC of 0.972 × 10−3 mm2 /s. The average ADC over the 17 cortical regions was 1.129 × 10−3 mm2 /s. When the average ADC value for the superior frontal gyrus (#2) at 0 degrees was compared with those at 20, 30, and 40 degrees, the P values were 0.028, 0.032, and 0.031, respectively. The differences of the average ADC values between the tilt angles were statistically significant (P < 0.05). For the other regions, however, there was no significant difference (P > 0.05) between the tilt angles. The relative differences of average ADC values between the tilt angles ranged from 0.7% to 12.2%, and the average of the relative differences was 3.5%.
Discussion As shown in Fig. 3a—e, the distortions in ADC maps differed from each other in the region close to the frontal sinus. The corresponding regions in the magnetic field distributions were changed by tilting the head of the digital phantom backwards (Fig. 1k—o). The main contribution of the magnetic field inhomogeneity due to the magnetic susceptibility effect was the magnetic susceptibility distribution of the object along the axis parallel to the static magnetic field [19]. When a tissue was placed over the region of air along the axis parallel to the static magnetic field, the magnetic field inhomogeneity occurred at regions around the air/tissue interface. This effect was dependent on the positional relationship between them on the axis parallel to the static field. When viewed from the perspective of the positional relationship on the axis parallel to the static field, the tissues placed over the frontal sinus were moved to the region over the soft tissues by tilting the head backwards (arrows in Fig. 1f—j). Since the inhomogeneities at the frontal region were reduced by this displacement (profiles shown in Fig. 1k—o), the distortions in ADC maps were reduced. This decrease of the inhomogeneity in the frontal region induced by tilting the head backwards was in agreement with the simulation results in a previous study [20]. These results suggest that the differences of the subject positioning cause the changes of the degree of distortions in EPI-based DW images as well as the inhomogeneity in human data. In our DW MRI simulator, we assume that each voxel contains a single tissue type, i.e., we do not take into account the partial volume effect within a voxel. In this study, the diffusion coefficients in GM and CSF were set to 0.87 × 10−3 mm2 /s and 3.34 ×10−3 mm2 /s, respectively. Therefore, ideally, the ADC value in GM becomes 0.87 × 10−3 mm2 /s. As shown in Table 1, however, the averages of the ADC values of 17 cortical regions were higher than that, and increasing to between 12% and 94%. This was thought to be due to signal contamination by CSF in the process from MR signal generation to image reconstruction. In the large distortion region (#2: superior frontal gyrus), there were statistically significant differences in the average ADC values between the different subject positions (P < 0.05). The range of the
154
S. Kumazawa et al.
Figure 3 a—e ADC maps at the slice level corresponding to the magnetic field distributions shown in Fig. 1f—j, respectively, f—j 17 cortical regions on the right cerebral hemisphere in each distorted ADC map.
relative differences in average ADC values by the differences in the subject position was approximately 1% to 12%. In our analysis based on the simulation, these variations in ADC values were dependent on the differences of the subject positioning, and did not include other factors such as image noise and region of interest measurements by the operator. The relative difference of average ADC values by the difference of the subject positioning was at the same level as the relative difference of those by the difference of MRI hardware (4%—9%) [23]. Although the partial volume effect was not taken into account in our simulation, voxels in the data
Table 1
gathered from actual human subjects contain a mixture of multiple tissues due to the low resolution of the DW image [15—17]. Since the ADC value in CSF is approximately three times larger than that in GM, in the cortical region adjacent to CSF, the ADC values are increased by the partial volume effect between GM and CSF [14]. Therefore, the variations of ADC values by the difference of the subject positioning in the cortical region in the data from actual human subjects could be larger than those demonstrated by our simulation. Our results suggest that the positioning of the subject against the static field is an additional factor affecting the
Averages and standard deviations of ADC values of the 17 cortical regions.
Cortical region #
0 degree
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
1.113 1.157 1.069 0.974 1.196 1.130 1.278 1.001 0.947 1.025 1.240 1.584 1.034 1.088 1.035 1.044 1.129
± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
10 degree 0.357 0.330 0.230 0.189 0.310 0.335 0.389 0.206 0.158 0.289 0.424 0.428 0.249 0.269 0.250 0.268 0.329
1.079 1.242 1.056 0.974 1.201 1.123 1.274 1.012 0.951 1.029 1.212 1.684 1.022 1.091 1.043 1.051 1.129
± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
0.312 0.367 0.233 0.187 0.314 0.329 0.387 0.212 0.163 0.302 0.404 0.518 0.231 0.278 0.260 0.274 0.332
20 degree 1.078 1.310 1.063 0.974 1.230 1.125 1.259 0.994 0.952 1.039 1.249 1.686 1.048 1.112 1.047 1.061 1.145
± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
0.320 0.397* 0.253 0.186 0.325 0.316 0.356 0.199 0.174 0.308 0.463 0.565 0.271 0.289 0.260 0.274 0.350
30 degree 1.078 1.311 1.054 0.979 1.225 1.105 1.301 0.988 0.956 1.028 1.244 1.618 1.017 1.080 1.019 1.038 1.122
± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
0.317 0.374* 0.248 0.193 0.337 0.312 0.402 0.180 0.188 0.310 0.443 0.448 0.243 0.264 0.255 0.276 0.314
40 degree 1.065 1.312 1.035 0.972 1.201 1.123 1.288 1.008 0.944 1.036 1.250 1.672 1.023 1.081 1.005 1.026 1.168
± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ± ±
0.297 0.395* 0.203 0.182 0.307 0.337 0.377 0.213 0.162 0.309 0.455 0.576 0.237 0.268 0.254 0.260 0.367
Relative differences [%] 4.4 12.2 3.2 0.7 2.8 2.2 3.2 2.4 1.3 1.4 3.1 6.2 3.0 2.9 4.1 3.4 4.0
The cortical region number corresponds to the number referred to in Fig. 2. Data are represented as the average ADC [10−3 mm2 /s] ± standard deviation. * P < 0.05 compared with the results at 0 degrees.
Changes of ADC due to subject positioning accuracy of the cortical ADC measurements. In order to investigate the effects of this factor for clinical practice, we plan to analyze the ADC values in the actual human subject data. Moreover, the signal contamination by CSF can also affect the measurements in WM region [14]. In our future work, we plan to investigate the changes of ADC value in WM region due to the positioning of the subject against a static field and the partial volume effect between GM and WM. In this study, we investigated the relationship between the cortical ADC values and the subject positioning by tilting the head of a digital phantom against a static field. As shown in Fig. 1e, a tilt angle of 40 degrees corresponds to the position with the subject’s chin up, and is within the range of potential adult head movement in the head coil. In the case of infants and children, the range of movement in the head coil is wider than that in adults. Therefore, to reduce the variation of the cortical ADC values, it is important to pay attention to the position of the subject’s head against the static field, as well as the reproducibility of this positioning.
Conclusion We demonstrated that the cortical ADC values could vary substantially according to changes in the positioning of the subject against a static field. Our results suggest that the positioning of the subject against a static field is one of the factors affecting the accuracy of the cortical ADC measurements.
Disclosure of interest The authors declare that they have no conflicts of interest concerning this article.
Acknowledgements This work was supported by Japan Society for the Promotion of Science (JSPS) KAKENHI Grant Number 23500561.
References [1] Le Bihan D, Johansen-Berg H. Diffusion MRI at 25: exploring brain tissue structure and function. Neuroimage 2012;61:324—41. [2] Le Bihan D. Apparent diffusion coefficient and beyond: what diffusion MR imaging can tell us about tissue structure. Radiology 2013;268:318—22. [3] Guo Y, Cai YQ, Cai ZL, et al. Differentiation of clinically benign and malignant breast lesions using diffusion-weighted imaging. J Magn Reson Imaging 2002;16:172—8. [4] Woodhams R, Matsunaga K, Iwabuchi K, et al. Diffusionweighted imaging of malignant breast tumors: the usefulness of apparent diffusion coeffi cient (ADC) value and ADC map for the detection of malignant breast tumors and evaluation of cancer extension. J Comput Assist Tomogr 2005;29:644—9. [5] Liu L, Yin B, Geng DY, et al. Comparison of ADC values of intracranial hemangiopericytomas and angiomatous and anaplastic meningiomas. J Neuroradiol 2014;41:188—94.
155 [6] von Kummer R, Dzialowski I, Gerber J. Therapeutic efficacy of brain imaging in acute ischemic stroke patients. J Neuroradiol 2014 [in press]. [7] Song I, Kim CK, Park BK, et al. Assessment of response to radiotherapy for prostate cancer: value of diffusion-weighted MRI at 3 T. AJR Am J Roentgenol 2010;194:W477—82. [8] Foltz WD, Wu A, Chung P, et al. Changes in apparent diffusion coefficient and T2 relaxation during radiotherapy for prostate cancer. J Magn Reson Imaging 2013;37:909—16. [9] Park SH, Moon WK, Cho N, et al. Diffusion-weighted MR imaging: pretreatment prediction of response to neoadjuvant chemotherapy in patients with breast cancer. Radiology 2010;257:56—63. [10] Bozzao A, Floris R, Baviera ME, et al. Diffusion and perfusion MR imaging in cases of Alzheimer’s disease: correlations with cortical atrophy and lesion load. AJNR Am J Neuroradiol 2001;22:1030—6. [11] Ray KM, Wang H, Chu Y, et al. Mild cognitive impairment: apparent diffusion coefficient in regional gray matter and white matter structures. Radiology 2006;241:197—205. [12] Liu T, Young G, Huang L, et al. 76-space analysis of grey matter diffusivity: methods and applications. Neuroimage 2006;31:51—65. [13] Le Bihan D, Poupon C, Amadon A, et al. Artifacts and pitfalls in diffusion MRI. J Magn Reson Imaging 2006;24:478—88. [14] Jones DK, Cercignani M. Twenty-five pitfalls in the analysis of diffusion MRI data. NMR Biomed 2010;23:803—20. [15] Yoshiura T, Mihara F, Tanaka A, et al. Novel method to estimate and display cerebral cortical degeneration using diffusion-weighted magnetic resonance imaging. Magn Reson Med 2005;54:455—9. [16] Kumazawa S, Yoshiura T, Honda H, et al. Partial volume estimation and segmentation of brain tissue based on diffusion tensor MRI. Med Phys 2010;37:1482—90. [17] Kumazawa S, Yoshiura T, Honda H, et al. Improvement of partial volume segmentation for brain tissue on diffusion tensor images using multiple-tensor estimation. J Digit Imaging 2013;26:1131—40. [18] Jezzard P, Balaban RS. Correction for geometric distortion in echo planar images from B0 field variations. Magn Reson Med 1995;34:65—73. [19] Yoder DA, Zhao Y, Paschal CB, et al. MRI simulator with object-specific field map calculations. Magn Reson Imaging 2004;22:315—28. [20] Truong TK, Clymer BD, Chakeres DW, Schmalbrock P. Threedimensional numerical simulations of susceptibility-induced magnetic field inhomogeneities in the human head. Magn Reson Imaging 2002;20:759—70. [21] Heberlein K, Hu X. Improved shim by subject head positioning. In: Proceedings of the ninth annual meeting of the international Society of Magnetic Resonance in medicine. 2001. p. 1157. [22] Dietrich O, Heiland S, Sartor K. Noise correction for the exact determination of apparent diffusion coefficients at low SNR. Magn Reson Med 2001;45:448—53. [23] Sasaki M, Yamada K, Watanabe Y, et al. Variability in absolute apparent diffusion coefficient values across different platforms may be substantial: a multivendor, multi-institutional comparison study. Radiology 2008;249:624—30. [24] Cocosco CA, Kollokian V, Kwan RK, Evans AC. BrainWeb: online interface to a 3D MRI simulated brain database. NeuroImage 1997;5:S425 [part 2/4]. [25] Tanner SF, Ramenghi LA, Ridgway JP, et al. Quantitative comparison of intrabrain diffusion in adults and preterm and term neonates and infants. AJR Am J Roentgenol 2000;174:1643—9.