- Email: [email protected]
Diffusion kurtosis imaging of the human kidney: A feasibility study
Magnetic Resonance Imaging 32 (2014) 413–420
Contents lists available at ScienceDirect
Magnetic Resonance Imaging journal homepage: www.mrijournal.com
Diffusion kurtosis imaging of the human kidney: A feasibility study☆,☆☆ Gael Pentang, Rotem Shlomo Lanzman, Philpp Heusch ⁎, Anja Müller-Lutz, Dirk Blondin, Gerald Antoch, Hans-Jörg Wittsack University Dusseldorf, Medical Faculty, Department of Diagnostic and Interventional Radiology, Moorenstrasse 5, D-40225 Düsseldorf, Germany
a r t i c l e
i n f o
Article history: Received 26 September 2013 Revised 9 January 2014 Accepted 14 January 2014 Keywords: Kurtosis Anisotropy Diffusion Kidney DKI DTI DWI
a b s t r a c t Purpose: To assess the feasibility and to optimize imaging parameters of diffusion kurtosis imaging (DKI) in human kidneys. Methods: The kidneys of ten healthy volunteers were examined on a clinical 3 T MR scanner. For DKI, respiratory triggered EPI sequences were acquired in the coronal plane (3 b-values: 0, 300, 600 s/mm 2, 30 diffusion directions). A goodness of fit analysis was performed and the influence of the signal-to-noise ratio (SNR) on the DKI results was evaluated. Region-of-interest (ROI) measurements were performed to determine apparent diffusion coefficient (ADC), fractional anisotropy (FA) and mean kurtosis (MK) of the cortex and the medulla of the kidneys. Intra-observer and inter-observer reproducibility using BlandAltman plots as well as subjective image quality of DKI were examined and ADC, FA, and MK parameters were compared. Results: The DKI model fitted better to the experimental data (r = 0.99) with p b 0.05 than the common mono-exponential ADC model (r = 0.96). Calculation of reliable kurtosis parameters in human kidneys requires a minimum SNR of 8.31 on b = 0 s/mm 2 images. Corticomedullary differentiation was possible on FA and MK maps. ADC, FA and MK revealed significant differences in medulla (ADC = 2.82 × 10 −3 mm2/s ± 0.25, FA = 0.42 ± 0. 05, MK = 0.78 ± 0.07) and cortex (ADC = 3.60 × 10−3 mm 2/s ± 0.28, FA = 0.18 ± 0.04, MK = 0.94 ± 0.07) with p b 0.001. Conclusion: Our initial results indicate the feasibility of DKI in the human kidney presuming an adequate SNR. Future studies in patients with kidney diseases are required to determine the value of DKI for functional kidney imaging. © 2014 Elsevier Inc. All rights reserved.
1. Introduction Diffusion tensor imaging (DTI) is a powerful method to assess the directionality of diffusion of water molecules in biological tissue known as Brownian motion [1]. It provides valuable information for the non-invasive characterization of tissue microstructural properties in vivo [2]. However, conventional DTI has limitations. Because of structural hindrances in biological tissue like membranes or directional structures as in renal medulla, the diffusion of water molecules is restricted and does not follow a Gaussian distribution. To describe the diffusion process more correctly, mathematical
☆ Conflicts of Interest and Source of Funding: None. ☆☆ Disclaimer: The concepts and information presented in this paper are based on research and are not commercially available. ⁎ Corresponding author at: University Dusseldorf, Department of Diagnostic and Interventional Radiology, Moorenstrasse 5, D-40225 Dusseldorf, Germany. Tel.: +49 211 81 17754; fax: +49 211 81 16299. E-mail address: [email protected] (P. Heusch). 0730-725X/$ – see front matter © 2014 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.mri.2014.01.006
models considering the deviation from the Gaussian behaviour have been proposed. Diffusional kurtosis imaging (DKI) is an extension of the conventional DTI model [3–8]. Along with the conventional diffusion tensor (DT), DKI estimates the kurtosis tensor (KT), which contains information about the deviation of the diffusion from the Gaussian form. DKI provides different diffusion parameters, such as kurtosis anisotropy (KA), mean kurtosis (MK), radial kurtosis (RK) and axial kurtosis (AK). DKI can better reflect the microstructural complexity of tissue because it considers the non-Gaussian behaviour of water in biological tissues. Consistently, Raab et al. have shown that DKI is superior to DTI for grading of cerebral gliomas [9]. A recent study of Falangola et al. [10] applied DKI for assessing aging-related changes in brain microstructure and showed a distinct signature for cerebrospinal fluid (CSF), grey matter (GM) and white matter (WM). DKI was also successfully applied for the detection of ischemic stroke and pathological changes in neural tissues as in Alzheimer disease [4]. Another major challenge of DKI in body imaging relates to the difficulty in obtaining sufficient SNR at high b-values. A parameter
414
G. Pentang et al. / Magnetic Resonance Imaging 32 (2014) 413–420
optimization is necessary to prevent low SNR, inherent in the diffusion technique, from impacting the result of key diffusion parameters [11]. Strategies that may be used to increase SNR include imaging at a higher field strength (3 T vs. 1.5 T); minimizing echo time (b 100 ms); increasing the number of signals acquired, which must be balanced against the resulting increase in imaging time. DKI has so far been applied to human and small animal brain studies. Non-Gaussian diffusion weighted imaging (DWI), not determining the complete kurtosis tensor, was used rarely in abdominal organs [12]. The human kidney is well suited for the application of DKI due the presence of anisotropy in renal tissue [13,14]. Thus the aim of this study was to assess the feasibility and reproducibility of DKI of the human kidney and to optimize imaging parameters. 2. Methods
model was used to fit the signal intensities S (see Eq. (1)) on a voxelby-voxel basis [4]. 3 3 Sðn; bÞ 1 2 2 3 3 3 3 ln ¼ −b ∑ ∑ ni nj Dij þ b D ∑ ∑ ∑ ∑ ni nj nk nl W ujkl ð1Þ S0 6 i¼1 j¼1 i¼1 j¼1 k¼1 l¼1 S(n,b) is the measured signal intensity for direction n depending on the diffusion-weighting value b, S0 is the signal intensity for b =0, Dij and Wijkl are the elements of the second order diffusion tensor (DT) DT and fourth-order KT W respectively, and D ¼ ð1=3Þtr ðDT Þ is mean diffusivity, where tr(DT) denotes the matrix trace. Diffusivity D (n) and kurtosis K(n) along direction n are given by 3
3
DðnÞ ¼ ∑ ∑ ni nj Diij ; i¼1 j¼1
ð2Þ
and 2
2.1. Study population The institutional review board approved our protocols and written informed consent was obtained from all volunteers before entering the study. Ten young healthy volunteers (6 men, 4 women, mean age 28.50 ± 3.34 years, range 28–34 years) without any history of renal disease, previous renal surgery, or any known systemic disease potentially involving the kidneys were included in this study. 2.2. Magnetic resonance imaging (MRI) MRI examinations were performed on a 3 T whole-body clinical MRI scanner (Magnetom Trio, a TIM system; Siemens Medical Systems, Erlangen, Germany) using a 6 channel array body coil and a 24 channel phased array spine coil integrated into the scanner table. For DKI, a single shot EPI sequence was applied in the coronal plane using respiratory triggering via a respiratory belt with 3 bvalues (0, 300 and 600 s/mm 2), 30 diffusion directions and 8 signal averages. The other imaging parameters were as follows: echo time (TE) = 90 ms, repetition time (TR) = 1500 ms, matrix = 192 × 192, field of view (FOV) = 400 mm, 10 slices with a slice thickness of 5 mm. GRAPPA (generalized autocalibrating partially parallel acquisition) as parallel imaging method was applied with an acceleration factor of 2. The mean acquisition time of the respiratory triggered DKI sequence was 32:08 ± 4:37 min (range, 23:56– 36:30 min). 2.3. Image analysis Initially, all diffusion-weighted (DW) images were reviewed including a subjective motion analysis by one of the authors, with more than 10 years’ experience in image processing and MR diffusion imaging, to assess whether the MR image quality was satisfactory for further analysis. For this purpose, a landmark was set on the first b = 0 s/mm 2 non DW image of the kidney of each volunteer that served as reference and then in the following b = 300, 600 s/mm 2 DW images. The displacement between the reference landmark and the landmarks in the DW images was measured to quantify motion. The results were averaged over all the subjects to obtain minimal and maximal values. All the acquired datasets were transferred to a workstation and processed to calculate the statistics of the diffusion and kurtosis parameters using in-house developed MATLAB algorithms based on [5,6]. A constrained linear least square formulation of the kurtosis
K ðnÞ ¼
3 3 3 3 D ∑ ∑ ∑ ∑ ni nj nk nl W ijkl : DðnÞ2 i¼1 j¼1 k¼1 l¼1
ð3Þ
Further, the conventional ADC determined by a logarithmic linear regression fit using the following equation was calculated [15]: 3 3 SðbÞ ¼ − ∑ ∑ bij Dij ln S0 i¼1 j¼1
ð4Þ
The estimated tensors were utilized to determine the diffusion and kurtosis measures for each subject corresponding to the parameters ADC, FA, MK according to the methods of Tabesh et al. [5] and Le Bihan et al. [15] respectively (see Eqs. (1) and (4)). Although values of RK and AK could be determined from our data, we chose not to report these in the present work, but either concentrate on investigating the relevance of MK measures for human kidney DKI. To optimize the DKI sequence by the means of acquisition time versus SNR, parametric images of ADC, FA and MK were calculated from subsets of the measured DWI including 2, 4, 6 and 8 signal averages. ROIs were manually drawn on the averaged b = 0 images for different signal averages. The b = 0 image was chosen for the measurements because of the lower SNR in the DW images [16]. SNR was calculated by dividing the mean signal intensity S within the ROI by the standard deviation (SD) of the background noise S SNR ¼ SD . Eight separate, manual drawn ROIs of 9 to 13 pixels were placed on FA maps because of its proven high corticomedullary discrimination [17]. The ROIs were drawn over the cortex and medulla on the upper pole, mid-zone and lower pole of the right kidney in each subject by one author experienced in genitourinary imaging and DTI measurements (Fig. 1). We selected the right kidney for analysis because of the relatively limited cardiac and respiratory motion artefact due to the presence of the liver above the right kidney [18]. The positions of the ROIs were reviewed by another author in consensus mode. These two authors were genitourinary radiologists with more than 5 years’ experience in MR imaging of the kidneys. ROIs on FA maps were copied onto the corresponding position on the ADC and MK maps. The mean and SD of the FA and MK values respectively as averaged values of the 4 ROIs on the cortex and the 4 ROIs on the medulla were calculated for all the signal averages to quantify the corticomedullary differentiation. As a marker of measurement error or reproducibility, the intraobserver and inter-observer variability of FA and MK values were examined using the free hand ROI technique [19]. Intra-observer repeatability analysis was based on data of one volunteer imaged on
G. Pentang et al. / Magnetic Resonance Imaging 32 (2014) 413–420
415
For qualitative evaluation, a 5-grade human observer study of MK and FA maps was conducted by the two radiologists taking into account the corticomedullary discrimination. Data of 6 subjects from all the signal averages (2, 4, 6, an 8) were included in this analysis. For graduation the following point-scale was used: 1, not evaluable; 2, poor (cortex-medulla difference is not visible); 3, moderate (cortex-medulla difference is visible but not clear); 4, good (reasonable cortex-medulla difference); 5, excellent (clear cortexmedulla difference). 2.4. Statistical analysis Statistical tests were performed using the curve fitting and statistics toolbox in MATLAB (Version: 8.0.0.783 (R2012b)). For all the tests a p-value of less than 0.05 was considered to indicate a statistically significant difference. To assess corticomedullary differentiation for various signal averaging (2, 4, 6 and 8 averages), a Student t test statistic was used. A goodness of fit evaluation was performed to test for the mathematical fitting of the mono-exponential and kurtosis models to the DWI data (see Eqs. (1) and (2)). The R 2 value was calculated, which is the square of the correlation between observed and expected outcome values. R 2 is expressed as Fig. 1. Free hand ROIs placed separately at the cortex and medulla of the upper pole, mid-zone and lower pole shown on the FA image. 2
R ¼ 1− different dates. The interval between the MRI measurements was 14 days. The FA and MK values of the first measurement were assessed and compared to the values obtained from the second measurement. The free hand ROIs used to measure the DKI metrics in the cortex and the medulla of the right kidney were placed by the one of the genitourinary radiologists mentioned above. For the inter-observer reproducibility, the two radiologists were instructed to independently place the ROIs on the cortex and medulla of the same volunteer using data of the first MRI measurement.
SSR SST
ð3Þ
where SSR is the sum of squares of the distance between the data points and the best-fit curve and SST is the sum of squares of the distances between the data points and the mean value of all data points. For the fitting, mean signal intensities of the ROIs placed over the cortex (see Fig. 1) of DWI images over all the volunteers for the 8 averages were computed. Then these averaged values were placed on a graph as function of the 3 b-values (0, 300 and 600 s/mm 2).
Fig. 2. EPI non diffusion-weighted image (b0), apparent diffusion coefficient (ADC), fractional anisotropy (FA), mean kurtosis (MK), axial kurtosis (AK) and radial kurtosis (RK) maps are shown for one coronal slice for one healthy volunteer.
416
G. Pentang et al. / Magnetic Resonance Imaging 32 (2014) 413–420
Table 1 Mean ± standard deviation (SD) of ADC, FA and MK values of the renal cortex and medulla for 2, 4, 6, 8 averages (av.) from data of 10 volunteers. No. of signal averages 2 av. 4 av. 6 av. 8 av.
ADC × 10−3 (mm2/s) Cortex 2.91 3.60 3.27 3.39
± ± ± ±
0.22 0.28 0.24 0.24
FA
MK
Medulla
Cortex
2.66 2.82 3.61 3.80
0.21 0.18 0.18 0.19
± ± ± ±
0.22 0.25 0.23 0.28
Repeated-measures analysis of variance (ANOVA) with the Tuckey’s honestly significant difference (HSD) post-hoc test was used to examine the effect of variable signal averaging on the cortexmedulla differences of MK. Intra- and inter-observer variability was calculated from reproducibility measurements using a two-sided, paired-samples Student’s t-test and Bland-Altman plots. The Bland-Altman plots show the agreement between two different measurements by plotting the difference versus the mean of the DKI parameters of the repeated measurements. This was done for the two measurements of the first observer and the first measurement of the first observer and the second observer. 3. Results In all subjects good image quality was achieved. A maximal subjective motion of 0.28 ± 0.02 mm (range, 0.05–0.29) was measured from the landmark displacement. Representative images are shown on Fig. 2. Mean ADC, FA and MK values were obtained with the four different signal averaging sequences. ADC values ranged from 2.91 × 10 −3 mm 2/s to 3.60 × 10 −3 mm 2/s in the cortex and from 2.66 × 10 −3 mm 2/s to 3.80 × 10 −3 mm 2/s in the medulla. FA of the cortex ranged from 0.18 to 0.21, whereas that of the medulla ranged from 0.38 to 0.46. MK of the renal cortex ranged from 0.91 to 0.94 and that of the medulla ranged from 0.74 to 0.86 (Table 1). MK values of the renal cortex were significantly higher than in the medulla while FA values, in the medulla were significantly higher than in the cortex (p b 0.001). Mean values for FA and MK are listed on Figs. 3 and 4. Respiratory triggered acquisitions with 4 averages (Total acquisition time was 15:47 ± 2:42 min, range: 13:26–19:06 min) exhibiting a SNR of 8.31 (p b 0.001) resulted in improved image quality with better corticomedullary differentiation in FA and MK maps compared to the sequence with 2 averages (SNR = 6.12). Whereas
± ± ± ±
0.05 0.04 0.03 0.03
Medulla
Cortex
0.38 0.42 0.46 0.43
0.93 0.94 0.91 0.91
± ± ± ±
0.06 0.05 0.10 0.07
± ± ± ±
Medulla 0.09 0.07 0.03 0.04
0.86 0.78 0.74 0.78
± ± ± ±
0.11 0.07 0.07 0.06
the use of 6 (SNR = 8.33) or 8 averages (SNR = 8.34) did not lead to a further improvement as compared to 4 averages. Representative images of one subject are shown in Fig. 5. The FA and MK measurements were again made twice by the first and once by the second observer using the free hand ROI technique including all the ROIs placed on the cortex and the medulla. The resulting values were compared using the two-sided, pairedsamples Student’s t-test. Respectively for the cortex and the medulla, the globally measured mean FAfirst/first was 0.19 ± 0.03 and 0.52 ± 0.04, the mean FAfirst/second was 0.20 ± 0.03 and 0.48 ± 0.07, and the mean FAsecond was 0.18 ± 0.02 and 0.47 ± 0.07. There was no significant difference between the first and second FA measurements of the first observer (p = 0.71; p = 0.29) and the measurements of the second observer (p = 0.73; p = 0.24). The globally measured mean MKfirst/first was 0.88 ± 0.10 and 0.74 ± 0.05, the mean MKfirst/second was 0.93 ± 0.05 and 0.79 ± 0.09, and the mean MKsecond was 0.85 ± 0.11 and 0.71 ± 0.06. There was no significant difference between the first and second FA measurements of the first observer (p = 0.13; p = 0.73) and the measurements of the second observer (p = 0.07; p = 0.47). The FA and MK measurements with their p-values are displayed in Tables 2 and 3. After the repeatability of the DKI parameters was reconfirmed, a qualitative evaluation of the FA and MK maps obtained from sequences with different number of averages (2, 4, 6, and 8) was performed. The sequence with 8 averages scored highest, followed by sequence with 6 and 4 averages; sequence with 2 averages had the lowest score (Table 4). Fig. 6 provides an example of the signal intensity decay curve of mean intensity of a ROI set on an homogeneous region in the kidney of one subject, as function of the b-value. The data points on the plot display a nonlinear decay. Also presented in this plot are the fitting curves from the mono-exponential fitting procedure and the diffusional kurtosis analysis. It can be clearly seen that the nonGaussian kurtosis analysis fits the data point substantially better
Fig. 3. Differences between the cortex (c) and medulla (m) on FA maps for the 2, 4, 6, 8 averages (av.). The y-axis reveals the different mean FA values between cortex and medulla over the 10 subjects.
G. Pentang et al. / Magnetic Resonance Imaging 32 (2014) 413–420
417
Fig. 4. Differences between the cortex (c) and medulla (m) on MK maps for the 2, 4, 6, 8 averages (av.). The y-axis reveals the different mean MK values between cortex and medulla over the 10 subjects.
than does the mono-exponential fitting procedure. The R 2 value for the mono-exponential fit was 0.96 and 0.99 for the kurtosis fit. As seen on the figure, a b-value of 600 s/mm 2 is already sufficient to observe the deviation of the renal MR diffusion signal from the mono-exponential behaviour. Repeated-measures analysis of variance by one-way ANOVA in MK showed that there was a statistically significant difference considering the cortex-medulla discrimination between groups with different signal averages (F(1,6) = 25.46, p = 0.0023). To further investigate the differences, a post-hoc analysis was performed with Tukey HSD using the sequence with 4 averages as control. Compared to the sequence with 2 averages, the sequence with 4 averages showed significantly higher cortex-medulla difference (p = 0.02). There were no statistically significant differences between sequences with 4, 6 and 8 averages. The results of the Bland-Altman analysis of the repeated measurements are shown in Fig. 7. This analysis showed that there is a good
agreement between the 1st and 2nd measurements of the 1st observer with − 0.02 as mean difference as well as between the 1st measurement of the 1st observer and the measurement of the 2nd observer with a mean difference of 0.03. There was no obvious deviation in relation to absolute values. All the recordings were located within the 95% limits of agreement. 4. Discussion In our study, we used respiratory triggered acquisitions to demonstrate that DKI of the kidneys is feasible, with good corticomedullary differentiation. DKI could be performed in all subjects with reliable image quality. Combining the results of the MK, FA and ADC-values, cortex-medulla contrast, reproducibility, and quantitative evaluation, a suitable DKI sequence should exhibit a SNR of 8.31 when b = 0, 300, 600 s/mm 2 are used. This was reached with our respiration triggered DKI sequence using 4 signal averages
Fig. 5. b = 0 images without diffusion weighting, ADC maps, FA maps and MK maps obtained with different sequences in the same volunteer with 2, 4, 6 and 8 averages (av.). The arrows point out better corticomedullary differentiation with 4, 6, 8 averages compared to sequence with 2 averages.
418
G. Pentang et al. / Magnetic Resonance Imaging 32 (2014) 413–420
Table 2 Mean ± SD values of FA and MK for each measurement (meas.) from the cortex and medulla for the two observers (obs.). Cortex
Medulla
FA 1st obs.1st meas. 1st obs. 2nd meas. 2nd obs. Mean
0.19 0.20 0.18 0.19
MK ± ± ± ±
0.03 0.03 0.02 0.03
0.88 0.93 0.85 0.88
FA ± ± ± ±
0.10 0.05 0.11 0.09
0.52 0.48 0.47 0.49
MK ± ± ± ±
0.04 0.07 0.07 0.06
0.74 0.79 0.71 0.75
± ± ± ±
0.05 0.09 0.06 0.14
and 30 diffusion-weighting directions which resulted in a total measurement time of about 15 min. Only few studies have reported the use of non-Gaussian DWI for abdominal organs. However, all these studies did not measure the complete kurtosis tensor. Rosenkrantz applied non-Gaussian DWI for a better characterization of diffusion processes in the prostate [20]. As an additional challenge, low SNR is a fundamental problem for calculation of parameters of higher diffusion models in abdominal organs [13]. From previous studies, the ADC of the normal kidney has been reported to have a spectrum from 3.00 to 1.5 × 10 −3 mm 2/s [19]. Thoeny et al.[21] demonstrated that using low b-values for DWI leads to no significant difference between ADC-values of the cortex and the medulla in healthy subjects. They attributed this result to the effect of higher true diffusion in the cortex being counteracted by greater anisotropy that stems from the radial orientation of medulla structures. Similar to our results, they reported a significant difference among ADC-values of the cortex and medulla with the use of higher b-values. The average FA values in our study were comparable to values reported by Ries et al.[22] but slightly lower than in other volunteer studies [23]. These differences might be explained by the influence of blood flow on diffusion coefficients [24]. Notohamiprodjo et al. [17] performed renal DTI using a 3 T system. In the healthy subjects, consistent with our results, the medulla was found to be more anisotropic with higher FA values than the cortex. They used two bvalues (200 and 400 s/mm 2) with 12 diffusion encoding directions and observed that the higher the b-values and the number of directions, the more accurate the measurement of diffusion is. Another study [25], reported that when similar acquisition time is maintained, a higher number of directions has greater effects on acquiring suitable DW images than increasing the number of averages and is much more important to get reliable diffusion indices. In addition, the departure of the diffusion process from the Gaussian model is well-observed with the use of higher b-values (1000, 2000 s/mm 2) in brain DKI [3–6,26]. In our study, we could show that b-values in the range of about 600–800 s/mm 2, are sufficient in abdominal DKI to observe the departure of the diffusion signal from mono-exponential behaviour. This was already demonstrated in the work of Wittsack et al. when evaluating the DWI signal of the human kidney with b-values up to 750 s/mm 2 [27]. Again, Rosenkrantz applied the kurtosis model in the prostate at a maximal diffusion strength of 800 s/mm 2 [20]. Because of the low SNR at high b-values in abdominal DKI and the above mentioned reasons, the Table 3 The p-values derived from a two-sided, paired-samples Student’s t-test. There was no significant difference between the first and second FA, MK measurements of the first observer and the measurements of the second observer. Cortex
Intra-observer Inter-observer
Medulla
FA
MK
FA
MK
0.71 0.73
0.13 0.07
0.29 0.24
0.73 0.47
Table 4 Qualitative evaluation of FA maps and MK maps from data of 6 volunteers. ⁎Evaluations were made in 5-grade scoring as follows: 1, not evaluable; 2, poor (cortex-medulla difference is not visible); 3, moderate (cortex-medulla difference is visible but not clear); 4, good (reasonable cortex-medulla difference); 5, excellent (clear cortex-medulla difference). No images included in the analysis were scored as 1. Evaluation⁎
2 av. 4 av. 6 av. 8av.
FA map
MK map
2
3
4
1
4 1
1 3 4 2
5
Mean
SD
2
3
0.63 0.75 0.52 0.52
3
2 2 4
3.00 4.17 4.33 4.67
3 5 4 4
4 1 1 1
5
Mean
SD
1 1
2.50 3.17 3.50 3.50
0.55 0.41 0.84 0.84
choice of 30 diffusion directions and b-values up to 600 s/mm 2 seems appropriate for renal DKI. To identify the parameters of the optimal sequence, we used corticomedullary differentiation on MK maps. Since the renal medulla is a radially oriented structure consisting of tubules, we expected differences in diffusion kurtosis as a directional measure. Previous studies reported a better characterization of tissue microstructure with kurtosis measurements in the brain [3,4,9,26]. Therefore one can expect that DKI parameters might differ between the renal cortex and medulla. Consistently MK of the cortex was constantly higher than that of the medulla in all four sequences. While the present study concentrates on the non-Gaussian analysis of the biological tissue microstructure using the kurtosis method, various groups did report results based on other higher diffusion models. As for example in a novel framework combining diffusion kurtosis and bi-exponential tensor analysis, Grinberg et al. [28] could highlight the non-Gaussian behaviour of water diffusion in human brain tissues using an extended range of b-factors (up to 7000 s/mm 2). However, this experimental method is not yet clinically feasible due to the long total acquisition time necessary to get sufficient SNR in particular in kidney imaging. Moreover, the bi-exponential behaviour of the diffusion MR signal in kidneys was shown before [26] and should be investigated in terms of tensorial analysis in future studies. In a recent study, Lanzman et al. could already highlight the potential of DT imaging for non-invasive functional assessment of transplanted kidneys [29]. They could show significant differences in FA values of the medulla between allograft recipients with heavily impaired renal function and those with moderate or mild impairment in renal function. Comparing MK values of normal kidneys with those of patients with various renal diseases may help to evaluate the clinical significance of renal kurtosis values and the role of the renal DKI. For instance in renal cancer, DKI may provide additional diagnostic information. Recently, Raab et al. [9] applied DK imaging in glioma and could differentiate between tumour grades using MK maps. Although the exact underlying meaning of the kurtosis findings has not yet been explained entirely these findings support the potential of DKI to reveal additional information to pathological alterations of the renal tissue. Since DKI has been proven to be more sensitive to tissue microstructure in comparison to ADC and FA measures, DKI of the kidney might be useful in evaluating conditions involving the renal tissue such as renal transplants, renal conditions after chemotherapy, or drug-induced renal tumours. Our study has some limitations. First, navigator-triggering was not possible due to technical limitations so that a respiratory belt had to be used. In clinical routine, the application of the respiratory belttype sensor may be suboptimal. Furthermore, respiratory movements of the kidney are mainly in a cranio-caudal direction, and do not always coincide with the abdominal wall movements. An approach that might further help reduce motion artifacts would be
G. Pentang et al. / Magnetic Resonance Imaging 32 (2014) 413–420
419
In conclusion, DKI of human kidneys is feasible. The use of 4 signal averages seems adequate in order to obtain good image quality when 3 b-values of (0, 300, 600 s/mm 2) and 30 diffusion directions are used.
Acknowledgments We thankfully acknowledge E. Rädisch for assistance with data acquisition.
References
Fig. 6. Example of a diffusion MR signal attenuation of the renal cortex (S/S0) against the b-value, including the mathematical fitting of the two models [1] and [2] to illustrate data modelling. The asterisks (*) represent the measured signal intensities within ROIs in the renal cortex (see Fig. 1) averaged over the 8 signal averages and all the volunteers; “mono-exp” and “kurt” denote the mono-exponential and the kurtosis model. The graph clearly illustrates the errors associated with the assumption of Gaussian distribution of water diffusion as in the case of the monoexponential fit (r = 0.96) versus a non-Gaussian distribution assumption from DKI (r = 0.99).
the use of navigator-echo type respiratory triggered acquisitions. They could help monitoring diaphragmatic motion and therefore decrease misregistration [25]. Due to technical difficulties on the MR scanner, this triggered acquisition technique could not be combined in our respiratory triggered DWI protocol. Second, the study was conducted on young healthy volunteers who were able to perform normal regular breathing. The results might differ in older subjects, patients in pain and patients who are less cooperative, having difficulties following a respiratory triggered acquisition. Further the hydration status of the kidney was not controlled in our study. It is known, that renal diffusion properties vary with water load. The influence of water load on the renal diffusion kurtosis still has to be investigated [17]. However, although we did not control the water load, our measurements of the reproducibility showed stable results of MK and FA within the error bounds. Further studies including patients with renal diseases should be conducted.
[1] Basser PJ, Mattiello J, LeBihan D. MR diffusion tensor spectroscopy and imaging. Biophys J 1994;66:259–67. [2] Basser PJ, Pierpaoli C. Microstructural and physiological features of tissues elucidated by quantitative-diffusion-tensor MRI. J Magn Reson B 1996;111: 209–19. [3] Jensen JH, Helpern JA, Ramani A, Lu H, Kaczynski K. Diffusional kurtosis imaging: the quantification of non-Gaussian water diffusion by means of magnetic resonance imaging. Magn Reson Med 2005;53:1432–40. [4] Jensen JH, Helpern JA. MRI quantification of non-Gaussian water diffusion by kurtosis analysis. NMR Biomed 2010;23:698–710. [5] Tabesh A, Jensen JH, Ardekani BA, Helpern JA. Estimation of tensors and tensorderived measures in diffusional kurtosis imaging. Magn Reson Med 2011;65: 823–36. [6] Tabesh A, Jensen JH, Ardekani BA, Helpern JA. Erratum: estimation of tensors and tensor-derived measures in diffusional kurtosis imaging. Magn Reson Med 2011;65:1507. [7] Cheung MM, Hui ES, Chan KC, Helpern JA, Qi L, Wu EX. Does diffusion kurtosis imaging lead to better neural tissue characterization? A rodent brain maturation study. NeuroImage 2009;45:386–92. [8] Hui ES, Cheung MM, Qi L, Wu EX. Towards better MR characterization of neural tissues using directional diffusion kurtosis analysis. NeuroImage 2008;42: 122–34. [9] Raab P, Hattingen E, Franz K, Zanella FE, Lanfermann H. Cerebral gliomas: diffusional kurtosis imaging analysis of microstructural differences. Radiology 2010;254:876–81. [10] Falangola MF, Jensen JH, Babb JS, Hu C, Castellanos FX, Di Martino A, et al. Agerelated non-Gaussian diffusion patterns in the prefrontal brain. J Magn Reson Imaging 2008;28:1345–50. [11] Qayyum A. Diffusion-weighted imaging in the abdomen and pelvis: concepts and applications. Radiographics 2009;29:1797–810. [12] Quentin M, Blondin D, Klasen J, Lanzman RS, Miese F-R, Arsov C, et al. Comparison of different mathematical models of diffusion-weighted prostate MR imaging. Magn Reson Imaging 2012;30:1468–74. [13] Yoshikawa T, Kawamitsu H, Mitchell DG, Ohno Y, Ku Y, Seo Y, et al. ADC measurement of abdominal organs and lesions using parallel imaging technique. AJR Am J Roentgenol 2006;187:1521–30. [14] Stejskal EO, Tanner JE. Spin diffusion measurements: spin echoes in the presence of a time-dependent field gradient. J Chem Phys 1965;42:288. [15] Denis Le Bihan, Mangin J-F, Poupon C, Clark CA, Pappata S, Molko N, et al. Diffusion Tensor Imaging: Concepts and Applications n.d. 2001. [16] Lagana M, Rovaris M, Ceccarelli A, Venturelli C, Marini S, Baselli G. DTI parameter optimisation for acquisition at 1.5 T: SNR analysis and clinical application. Comput Intell Neurosci 2010;2010.
Fig. 7. Bland-Altman analysis of the difference between the repeated measurements of the two observers.
420
G. Pentang et al. / Magnetic Resonance Imaging 32 (2014) 413–420
[17] Notohamiprodjo M, Dietrich O, Horger W, Horng A, Helck AD, Herrmann KA, et al. Diffusion tensor imaging (DTI) of the kidney at 3 Tesla – feasibility, protocol evaluation and comparison to 1.5 Tesla. Invest Radiol 2010;45:245–54. [18] Kido A, Kataoka M, Yamamoto A, Nakamoto Y, Umeoka S, Koyama T, et al. Diffusion tensor MRI of the kidney at 3.0 and 1.5 Tesla. Acta Radiol Stockh Swed 1987 2010;51:1059–63. [19] Gurses B, Kilickesmez O, Tasdelen N, Firat Z, Gurmen N. Diffusion tensor imaging of the kidney at 3 tesla: normative values and repeatability of measurements in healthy volunteers. Diagn Interv Radiol 2011;17(4):317–22. [20] Rosenkrantz AB, Sigmund EE, Johnson G, Babb JS, Mussi TC, Melamed J, et al. Prostate cancer: feasibility and preliminary experience of a diffusional kurtosis model for detection and assessment of aggressiveness of peripheral zone cancer. Radiology 2012;264:126–35. [21] Thoeny HC, Keyzer FD, Oyen RH, Peeters RR. Diffusion-weighted MR imaging of kidneys in healthy volunteers and patients with parenchymal diseases: initial Experience1. Radiology 2005;235:911–7. [22] Ries M, Jones RA, Basseau F, Moonen CTW, Grenier N. Diffusion tensor MRI of the human kidney. J Magn Reson Imaging 2001;14:42–9. [23] Müller MF, Prasad PV, Bimmler D, Kaiser A, Edelman RR. Functional imaging of the kidney by means of measurement of the apparent diffusion coefficient. Radiology 1994;193:711–5.
[24] Heusch P, Wittsack H-J, Kröpil P, Blondin D, Quentin M, Klasen J, et al. Impact of blood flow on diffusion coefficients of the human kidney: a time-resolved ECG-triggered diffusion-tensor imaging (DTI) study at 3 T. J Magn Reson Imaging 2013;37:233–6. [25] Kataoka M, Kido A, Yamamoto A, Nakamoto Y, Koyama T, Isoda H, et al. Diffusion tensor imaging of kidneys with respiratory triggering: optimization of parameters to demonstrate anisotropic structures on fraction anisotropy maps. J Magn Reson Imaging 2009;29:736–44. [26] Jansen JFA, Stambuk HE, Koutcher JA, Shukla-Dave A. Non-Gaussian analysis of diffusion-weighted MR imaging in head and neck squamous cell carcinoma: a feasibility study. Am J Neuroradiol 2010;31:741–8. [27] Wittsack H-J, Lanzman RS, Mathys C, Janssen H, Mödder U, Blondin D. Statistical evaluation of diffusion-weighted imaging of the human kidney. Magn Reson Med 2010;64:616–22. [28] Grinberg F, Farrher E, Kaffanke J, Oros-Peusquens A-M, Shah NJ. Non-Gaussian diffusion in human brain tissue at high b-factors as examined by a combined diffusion kurtosis and biexponential diffusion tensor analysis. NeuroImage 2011;57:1087–102. [29] Lanzman RS, Ljimani A, Pentang G, Zgoura P, Zenginli H, Kröpil P, et al. Kidney transplant: functional assessment with diffusion-tensor MR imaging at 3 T. Radiology 2013;266:218–25.
Contents lists available at ScienceDirect
Magnetic Resonance Imaging journal homepage: www.mrijournal.com
Diffusion kurtosis imaging of the human kidney: A feasibility study☆,☆☆ Gael Pentang, Rotem Shlomo Lanzman, Philpp Heusch ⁎, Anja Müller-Lutz, Dirk Blondin, Gerald Antoch, Hans-Jörg Wittsack University Dusseldorf, Medical Faculty, Department of Diagnostic and Interventional Radiology, Moorenstrasse 5, D-40225 Düsseldorf, Germany
a r t i c l e
i n f o
Article history: Received 26 September 2013 Revised 9 January 2014 Accepted 14 January 2014 Keywords: Kurtosis Anisotropy Diffusion Kidney DKI DTI DWI
a b s t r a c t Purpose: To assess the feasibility and to optimize imaging parameters of diffusion kurtosis imaging (DKI) in human kidneys. Methods: The kidneys of ten healthy volunteers were examined on a clinical 3 T MR scanner. For DKI, respiratory triggered EPI sequences were acquired in the coronal plane (3 b-values: 0, 300, 600 s/mm 2, 30 diffusion directions). A goodness of fit analysis was performed and the influence of the signal-to-noise ratio (SNR) on the DKI results was evaluated. Region-of-interest (ROI) measurements were performed to determine apparent diffusion coefficient (ADC), fractional anisotropy (FA) and mean kurtosis (MK) of the cortex and the medulla of the kidneys. Intra-observer and inter-observer reproducibility using BlandAltman plots as well as subjective image quality of DKI were examined and ADC, FA, and MK parameters were compared. Results: The DKI model fitted better to the experimental data (r = 0.99) with p b 0.05 than the common mono-exponential ADC model (r = 0.96). Calculation of reliable kurtosis parameters in human kidneys requires a minimum SNR of 8.31 on b = 0 s/mm 2 images. Corticomedullary differentiation was possible on FA and MK maps. ADC, FA and MK revealed significant differences in medulla (ADC = 2.82 × 10 −3 mm2/s ± 0.25, FA = 0.42 ± 0. 05, MK = 0.78 ± 0.07) and cortex (ADC = 3.60 × 10−3 mm 2/s ± 0.28, FA = 0.18 ± 0.04, MK = 0.94 ± 0.07) with p b 0.001. Conclusion: Our initial results indicate the feasibility of DKI in the human kidney presuming an adequate SNR. Future studies in patients with kidney diseases are required to determine the value of DKI for functional kidney imaging. © 2014 Elsevier Inc. All rights reserved.
1. Introduction Diffusion tensor imaging (DTI) is a powerful method to assess the directionality of diffusion of water molecules in biological tissue known as Brownian motion [1]. It provides valuable information for the non-invasive characterization of tissue microstructural properties in vivo [2]. However, conventional DTI has limitations. Because of structural hindrances in biological tissue like membranes or directional structures as in renal medulla, the diffusion of water molecules is restricted and does not follow a Gaussian distribution. To describe the diffusion process more correctly, mathematical
☆ Conflicts of Interest and Source of Funding: None. ☆☆ Disclaimer: The concepts and information presented in this paper are based on research and are not commercially available. ⁎ Corresponding author at: University Dusseldorf, Department of Diagnostic and Interventional Radiology, Moorenstrasse 5, D-40225 Dusseldorf, Germany. Tel.: +49 211 81 17754; fax: +49 211 81 16299. E-mail address: [email protected] (P. Heusch). 0730-725X/$ – see front matter © 2014 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.mri.2014.01.006
models considering the deviation from the Gaussian behaviour have been proposed. Diffusional kurtosis imaging (DKI) is an extension of the conventional DTI model [3–8]. Along with the conventional diffusion tensor (DT), DKI estimates the kurtosis tensor (KT), which contains information about the deviation of the diffusion from the Gaussian form. DKI provides different diffusion parameters, such as kurtosis anisotropy (KA), mean kurtosis (MK), radial kurtosis (RK) and axial kurtosis (AK). DKI can better reflect the microstructural complexity of tissue because it considers the non-Gaussian behaviour of water in biological tissues. Consistently, Raab et al. have shown that DKI is superior to DTI for grading of cerebral gliomas [9]. A recent study of Falangola et al. [10] applied DKI for assessing aging-related changes in brain microstructure and showed a distinct signature for cerebrospinal fluid (CSF), grey matter (GM) and white matter (WM). DKI was also successfully applied for the detection of ischemic stroke and pathological changes in neural tissues as in Alzheimer disease [4]. Another major challenge of DKI in body imaging relates to the difficulty in obtaining sufficient SNR at high b-values. A parameter
414
G. Pentang et al. / Magnetic Resonance Imaging 32 (2014) 413–420
optimization is necessary to prevent low SNR, inherent in the diffusion technique, from impacting the result of key diffusion parameters [11]. Strategies that may be used to increase SNR include imaging at a higher field strength (3 T vs. 1.5 T); minimizing echo time (b 100 ms); increasing the number of signals acquired, which must be balanced against the resulting increase in imaging time. DKI has so far been applied to human and small animal brain studies. Non-Gaussian diffusion weighted imaging (DWI), not determining the complete kurtosis tensor, was used rarely in abdominal organs [12]. The human kidney is well suited for the application of DKI due the presence of anisotropy in renal tissue [13,14]. Thus the aim of this study was to assess the feasibility and reproducibility of DKI of the human kidney and to optimize imaging parameters. 2. Methods
model was used to fit the signal intensities S (see Eq. (1)) on a voxelby-voxel basis [4]. 3 3 Sðn; bÞ 1 2 2 3 3 3 3 ln ¼ −b ∑ ∑ ni nj Dij þ b D ∑ ∑ ∑ ∑ ni nj nk nl W ujkl ð1Þ S0 6 i¼1 j¼1 i¼1 j¼1 k¼1 l¼1 S(n,b) is the measured signal intensity for direction n depending on the diffusion-weighting value b, S0 is the signal intensity for b =0, Dij and Wijkl are the elements of the second order diffusion tensor (DT) DT and fourth-order KT W respectively, and D ¼ ð1=3Þtr ðDT Þ is mean diffusivity, where tr(DT) denotes the matrix trace. Diffusivity D (n) and kurtosis K(n) along direction n are given by 3
3
DðnÞ ¼ ∑ ∑ ni nj Diij ; i¼1 j¼1
ð2Þ
and 2
2.1. Study population The institutional review board approved our protocols and written informed consent was obtained from all volunteers before entering the study. Ten young healthy volunteers (6 men, 4 women, mean age 28.50 ± 3.34 years, range 28–34 years) without any history of renal disease, previous renal surgery, or any known systemic disease potentially involving the kidneys were included in this study. 2.2. Magnetic resonance imaging (MRI) MRI examinations were performed on a 3 T whole-body clinical MRI scanner (Magnetom Trio, a TIM system; Siemens Medical Systems, Erlangen, Germany) using a 6 channel array body coil and a 24 channel phased array spine coil integrated into the scanner table. For DKI, a single shot EPI sequence was applied in the coronal plane using respiratory triggering via a respiratory belt with 3 bvalues (0, 300 and 600 s/mm 2), 30 diffusion directions and 8 signal averages. The other imaging parameters were as follows: echo time (TE) = 90 ms, repetition time (TR) = 1500 ms, matrix = 192 × 192, field of view (FOV) = 400 mm, 10 slices with a slice thickness of 5 mm. GRAPPA (generalized autocalibrating partially parallel acquisition) as parallel imaging method was applied with an acceleration factor of 2. The mean acquisition time of the respiratory triggered DKI sequence was 32:08 ± 4:37 min (range, 23:56– 36:30 min). 2.3. Image analysis Initially, all diffusion-weighted (DW) images were reviewed including a subjective motion analysis by one of the authors, with more than 10 years’ experience in image processing and MR diffusion imaging, to assess whether the MR image quality was satisfactory for further analysis. For this purpose, a landmark was set on the first b = 0 s/mm 2 non DW image of the kidney of each volunteer that served as reference and then in the following b = 300, 600 s/mm 2 DW images. The displacement between the reference landmark and the landmarks in the DW images was measured to quantify motion. The results were averaged over all the subjects to obtain minimal and maximal values. All the acquired datasets were transferred to a workstation and processed to calculate the statistics of the diffusion and kurtosis parameters using in-house developed MATLAB algorithms based on [5,6]. A constrained linear least square formulation of the kurtosis
K ðnÞ ¼
3 3 3 3 D ∑ ∑ ∑ ∑ ni nj nk nl W ijkl : DðnÞ2 i¼1 j¼1 k¼1 l¼1
ð3Þ
Further, the conventional ADC determined by a logarithmic linear regression fit using the following equation was calculated [15]: 3 3 SðbÞ ¼ − ∑ ∑ bij Dij ln S0 i¼1 j¼1
ð4Þ
The estimated tensors were utilized to determine the diffusion and kurtosis measures for each subject corresponding to the parameters ADC, FA, MK according to the methods of Tabesh et al. [5] and Le Bihan et al. [15] respectively (see Eqs. (1) and (4)). Although values of RK and AK could be determined from our data, we chose not to report these in the present work, but either concentrate on investigating the relevance of MK measures for human kidney DKI. To optimize the DKI sequence by the means of acquisition time versus SNR, parametric images of ADC, FA and MK were calculated from subsets of the measured DWI including 2, 4, 6 and 8 signal averages. ROIs were manually drawn on the averaged b = 0 images for different signal averages. The b = 0 image was chosen for the measurements because of the lower SNR in the DW images [16]. SNR was calculated by dividing the mean signal intensity S within the ROI by the standard deviation (SD) of the background noise S SNR ¼ SD . Eight separate, manual drawn ROIs of 9 to 13 pixels were placed on FA maps because of its proven high corticomedullary discrimination [17]. The ROIs were drawn over the cortex and medulla on the upper pole, mid-zone and lower pole of the right kidney in each subject by one author experienced in genitourinary imaging and DTI measurements (Fig. 1). We selected the right kidney for analysis because of the relatively limited cardiac and respiratory motion artefact due to the presence of the liver above the right kidney [18]. The positions of the ROIs were reviewed by another author in consensus mode. These two authors were genitourinary radiologists with more than 5 years’ experience in MR imaging of the kidneys. ROIs on FA maps were copied onto the corresponding position on the ADC and MK maps. The mean and SD of the FA and MK values respectively as averaged values of the 4 ROIs on the cortex and the 4 ROIs on the medulla were calculated for all the signal averages to quantify the corticomedullary differentiation. As a marker of measurement error or reproducibility, the intraobserver and inter-observer variability of FA and MK values were examined using the free hand ROI technique [19]. Intra-observer repeatability analysis was based on data of one volunteer imaged on
G. Pentang et al. / Magnetic Resonance Imaging 32 (2014) 413–420
415
For qualitative evaluation, a 5-grade human observer study of MK and FA maps was conducted by the two radiologists taking into account the corticomedullary discrimination. Data of 6 subjects from all the signal averages (2, 4, 6, an 8) were included in this analysis. For graduation the following point-scale was used: 1, not evaluable; 2, poor (cortex-medulla difference is not visible); 3, moderate (cortex-medulla difference is visible but not clear); 4, good (reasonable cortex-medulla difference); 5, excellent (clear cortexmedulla difference). 2.4. Statistical analysis Statistical tests were performed using the curve fitting and statistics toolbox in MATLAB (Version: 8.0.0.783 (R2012b)). For all the tests a p-value of less than 0.05 was considered to indicate a statistically significant difference. To assess corticomedullary differentiation for various signal averaging (2, 4, 6 and 8 averages), a Student t test statistic was used. A goodness of fit evaluation was performed to test for the mathematical fitting of the mono-exponential and kurtosis models to the DWI data (see Eqs. (1) and (2)). The R 2 value was calculated, which is the square of the correlation between observed and expected outcome values. R 2 is expressed as Fig. 1. Free hand ROIs placed separately at the cortex and medulla of the upper pole, mid-zone and lower pole shown on the FA image. 2
R ¼ 1− different dates. The interval between the MRI measurements was 14 days. The FA and MK values of the first measurement were assessed and compared to the values obtained from the second measurement. The free hand ROIs used to measure the DKI metrics in the cortex and the medulla of the right kidney were placed by the one of the genitourinary radiologists mentioned above. For the inter-observer reproducibility, the two radiologists were instructed to independently place the ROIs on the cortex and medulla of the same volunteer using data of the first MRI measurement.
SSR SST
ð3Þ
where SSR is the sum of squares of the distance between the data points and the best-fit curve and SST is the sum of squares of the distances between the data points and the mean value of all data points. For the fitting, mean signal intensities of the ROIs placed over the cortex (see Fig. 1) of DWI images over all the volunteers for the 8 averages were computed. Then these averaged values were placed on a graph as function of the 3 b-values (0, 300 and 600 s/mm 2).
Fig. 2. EPI non diffusion-weighted image (b0), apparent diffusion coefficient (ADC), fractional anisotropy (FA), mean kurtosis (MK), axial kurtosis (AK) and radial kurtosis (RK) maps are shown for one coronal slice for one healthy volunteer.
416
G. Pentang et al. / Magnetic Resonance Imaging 32 (2014) 413–420
Table 1 Mean ± standard deviation (SD) of ADC, FA and MK values of the renal cortex and medulla for 2, 4, 6, 8 averages (av.) from data of 10 volunteers. No. of signal averages 2 av. 4 av. 6 av. 8 av.
ADC × 10−3 (mm2/s) Cortex 2.91 3.60 3.27 3.39
± ± ± ±
0.22 0.28 0.24 0.24
FA
MK
Medulla
Cortex
2.66 2.82 3.61 3.80
0.21 0.18 0.18 0.19
± ± ± ±
0.22 0.25 0.23 0.28
Repeated-measures analysis of variance (ANOVA) with the Tuckey’s honestly significant difference (HSD) post-hoc test was used to examine the effect of variable signal averaging on the cortexmedulla differences of MK. Intra- and inter-observer variability was calculated from reproducibility measurements using a two-sided, paired-samples Student’s t-test and Bland-Altman plots. The Bland-Altman plots show the agreement between two different measurements by plotting the difference versus the mean of the DKI parameters of the repeated measurements. This was done for the two measurements of the first observer and the first measurement of the first observer and the second observer. 3. Results In all subjects good image quality was achieved. A maximal subjective motion of 0.28 ± 0.02 mm (range, 0.05–0.29) was measured from the landmark displacement. Representative images are shown on Fig. 2. Mean ADC, FA and MK values were obtained with the four different signal averaging sequences. ADC values ranged from 2.91 × 10 −3 mm 2/s to 3.60 × 10 −3 mm 2/s in the cortex and from 2.66 × 10 −3 mm 2/s to 3.80 × 10 −3 mm 2/s in the medulla. FA of the cortex ranged from 0.18 to 0.21, whereas that of the medulla ranged from 0.38 to 0.46. MK of the renal cortex ranged from 0.91 to 0.94 and that of the medulla ranged from 0.74 to 0.86 (Table 1). MK values of the renal cortex were significantly higher than in the medulla while FA values, in the medulla were significantly higher than in the cortex (p b 0.001). Mean values for FA and MK are listed on Figs. 3 and 4. Respiratory triggered acquisitions with 4 averages (Total acquisition time was 15:47 ± 2:42 min, range: 13:26–19:06 min) exhibiting a SNR of 8.31 (p b 0.001) resulted in improved image quality with better corticomedullary differentiation in FA and MK maps compared to the sequence with 2 averages (SNR = 6.12). Whereas
± ± ± ±
0.05 0.04 0.03 0.03
Medulla
Cortex
0.38 0.42 0.46 0.43
0.93 0.94 0.91 0.91
± ± ± ±
0.06 0.05 0.10 0.07
± ± ± ±
Medulla 0.09 0.07 0.03 0.04
0.86 0.78 0.74 0.78
± ± ± ±
0.11 0.07 0.07 0.06
the use of 6 (SNR = 8.33) or 8 averages (SNR = 8.34) did not lead to a further improvement as compared to 4 averages. Representative images of one subject are shown in Fig. 5. The FA and MK measurements were again made twice by the first and once by the second observer using the free hand ROI technique including all the ROIs placed on the cortex and the medulla. The resulting values were compared using the two-sided, pairedsamples Student’s t-test. Respectively for the cortex and the medulla, the globally measured mean FAfirst/first was 0.19 ± 0.03 and 0.52 ± 0.04, the mean FAfirst/second was 0.20 ± 0.03 and 0.48 ± 0.07, and the mean FAsecond was 0.18 ± 0.02 and 0.47 ± 0.07. There was no significant difference between the first and second FA measurements of the first observer (p = 0.71; p = 0.29) and the measurements of the second observer (p = 0.73; p = 0.24). The globally measured mean MKfirst/first was 0.88 ± 0.10 and 0.74 ± 0.05, the mean MKfirst/second was 0.93 ± 0.05 and 0.79 ± 0.09, and the mean MKsecond was 0.85 ± 0.11 and 0.71 ± 0.06. There was no significant difference between the first and second FA measurements of the first observer (p = 0.13; p = 0.73) and the measurements of the second observer (p = 0.07; p = 0.47). The FA and MK measurements with their p-values are displayed in Tables 2 and 3. After the repeatability of the DKI parameters was reconfirmed, a qualitative evaluation of the FA and MK maps obtained from sequences with different number of averages (2, 4, 6, and 8) was performed. The sequence with 8 averages scored highest, followed by sequence with 6 and 4 averages; sequence with 2 averages had the lowest score (Table 4). Fig. 6 provides an example of the signal intensity decay curve of mean intensity of a ROI set on an homogeneous region in the kidney of one subject, as function of the b-value. The data points on the plot display a nonlinear decay. Also presented in this plot are the fitting curves from the mono-exponential fitting procedure and the diffusional kurtosis analysis. It can be clearly seen that the nonGaussian kurtosis analysis fits the data point substantially better
Fig. 3. Differences between the cortex (c) and medulla (m) on FA maps for the 2, 4, 6, 8 averages (av.). The y-axis reveals the different mean FA values between cortex and medulla over the 10 subjects.
G. Pentang et al. / Magnetic Resonance Imaging 32 (2014) 413–420
417
Fig. 4. Differences between the cortex (c) and medulla (m) on MK maps for the 2, 4, 6, 8 averages (av.). The y-axis reveals the different mean MK values between cortex and medulla over the 10 subjects.
than does the mono-exponential fitting procedure. The R 2 value for the mono-exponential fit was 0.96 and 0.99 for the kurtosis fit. As seen on the figure, a b-value of 600 s/mm 2 is already sufficient to observe the deviation of the renal MR diffusion signal from the mono-exponential behaviour. Repeated-measures analysis of variance by one-way ANOVA in MK showed that there was a statistically significant difference considering the cortex-medulla discrimination between groups with different signal averages (F(1,6) = 25.46, p = 0.0023). To further investigate the differences, a post-hoc analysis was performed with Tukey HSD using the sequence with 4 averages as control. Compared to the sequence with 2 averages, the sequence with 4 averages showed significantly higher cortex-medulla difference (p = 0.02). There were no statistically significant differences between sequences with 4, 6 and 8 averages. The results of the Bland-Altman analysis of the repeated measurements are shown in Fig. 7. This analysis showed that there is a good
agreement between the 1st and 2nd measurements of the 1st observer with − 0.02 as mean difference as well as between the 1st measurement of the 1st observer and the measurement of the 2nd observer with a mean difference of 0.03. There was no obvious deviation in relation to absolute values. All the recordings were located within the 95% limits of agreement. 4. Discussion In our study, we used respiratory triggered acquisitions to demonstrate that DKI of the kidneys is feasible, with good corticomedullary differentiation. DKI could be performed in all subjects with reliable image quality. Combining the results of the MK, FA and ADC-values, cortex-medulla contrast, reproducibility, and quantitative evaluation, a suitable DKI sequence should exhibit a SNR of 8.31 when b = 0, 300, 600 s/mm 2 are used. This was reached with our respiration triggered DKI sequence using 4 signal averages
Fig. 5. b = 0 images without diffusion weighting, ADC maps, FA maps and MK maps obtained with different sequences in the same volunteer with 2, 4, 6 and 8 averages (av.). The arrows point out better corticomedullary differentiation with 4, 6, 8 averages compared to sequence with 2 averages.
418
G. Pentang et al. / Magnetic Resonance Imaging 32 (2014) 413–420
Table 2 Mean ± SD values of FA and MK for each measurement (meas.) from the cortex and medulla for the two observers (obs.). Cortex
Medulla
FA 1st obs.1st meas. 1st obs. 2nd meas. 2nd obs. Mean
0.19 0.20 0.18 0.19
MK ± ± ± ±
0.03 0.03 0.02 0.03
0.88 0.93 0.85 0.88
FA ± ± ± ±
0.10 0.05 0.11 0.09
0.52 0.48 0.47 0.49
MK ± ± ± ±
0.04 0.07 0.07 0.06
0.74 0.79 0.71 0.75
± ± ± ±
0.05 0.09 0.06 0.14
and 30 diffusion-weighting directions which resulted in a total measurement time of about 15 min. Only few studies have reported the use of non-Gaussian DWI for abdominal organs. However, all these studies did not measure the complete kurtosis tensor. Rosenkrantz applied non-Gaussian DWI for a better characterization of diffusion processes in the prostate [20]. As an additional challenge, low SNR is a fundamental problem for calculation of parameters of higher diffusion models in abdominal organs [13]. From previous studies, the ADC of the normal kidney has been reported to have a spectrum from 3.00 to 1.5 × 10 −3 mm 2/s [19]. Thoeny et al.[21] demonstrated that using low b-values for DWI leads to no significant difference between ADC-values of the cortex and the medulla in healthy subjects. They attributed this result to the effect of higher true diffusion in the cortex being counteracted by greater anisotropy that stems from the radial orientation of medulla structures. Similar to our results, they reported a significant difference among ADC-values of the cortex and medulla with the use of higher b-values. The average FA values in our study were comparable to values reported by Ries et al.[22] but slightly lower than in other volunteer studies [23]. These differences might be explained by the influence of blood flow on diffusion coefficients [24]. Notohamiprodjo et al. [17] performed renal DTI using a 3 T system. In the healthy subjects, consistent with our results, the medulla was found to be more anisotropic with higher FA values than the cortex. They used two bvalues (200 and 400 s/mm 2) with 12 diffusion encoding directions and observed that the higher the b-values and the number of directions, the more accurate the measurement of diffusion is. Another study [25], reported that when similar acquisition time is maintained, a higher number of directions has greater effects on acquiring suitable DW images than increasing the number of averages and is much more important to get reliable diffusion indices. In addition, the departure of the diffusion process from the Gaussian model is well-observed with the use of higher b-values (1000, 2000 s/mm 2) in brain DKI [3–6,26]. In our study, we could show that b-values in the range of about 600–800 s/mm 2, are sufficient in abdominal DKI to observe the departure of the diffusion signal from mono-exponential behaviour. This was already demonstrated in the work of Wittsack et al. when evaluating the DWI signal of the human kidney with b-values up to 750 s/mm 2 [27]. Again, Rosenkrantz applied the kurtosis model in the prostate at a maximal diffusion strength of 800 s/mm 2 [20]. Because of the low SNR at high b-values in abdominal DKI and the above mentioned reasons, the Table 3 The p-values derived from a two-sided, paired-samples Student’s t-test. There was no significant difference between the first and second FA, MK measurements of the first observer and the measurements of the second observer. Cortex
Intra-observer Inter-observer
Medulla
FA
MK
FA
MK
0.71 0.73
0.13 0.07
0.29 0.24
0.73 0.47
Table 4 Qualitative evaluation of FA maps and MK maps from data of 6 volunteers. ⁎Evaluations were made in 5-grade scoring as follows: 1, not evaluable; 2, poor (cortex-medulla difference is not visible); 3, moderate (cortex-medulla difference is visible but not clear); 4, good (reasonable cortex-medulla difference); 5, excellent (clear cortex-medulla difference). No images included in the analysis were scored as 1. Evaluation⁎
2 av. 4 av. 6 av. 8av.
FA map
MK map
2
3
4
1
4 1
1 3 4 2
5
Mean
SD
2
3
0.63 0.75 0.52 0.52
3
2 2 4
3.00 4.17 4.33 4.67
3 5 4 4
4 1 1 1
5
Mean
SD
1 1
2.50 3.17 3.50 3.50
0.55 0.41 0.84 0.84
choice of 30 diffusion directions and b-values up to 600 s/mm 2 seems appropriate for renal DKI. To identify the parameters of the optimal sequence, we used corticomedullary differentiation on MK maps. Since the renal medulla is a radially oriented structure consisting of tubules, we expected differences in diffusion kurtosis as a directional measure. Previous studies reported a better characterization of tissue microstructure with kurtosis measurements in the brain [3,4,9,26]. Therefore one can expect that DKI parameters might differ between the renal cortex and medulla. Consistently MK of the cortex was constantly higher than that of the medulla in all four sequences. While the present study concentrates on the non-Gaussian analysis of the biological tissue microstructure using the kurtosis method, various groups did report results based on other higher diffusion models. As for example in a novel framework combining diffusion kurtosis and bi-exponential tensor analysis, Grinberg et al. [28] could highlight the non-Gaussian behaviour of water diffusion in human brain tissues using an extended range of b-factors (up to 7000 s/mm 2). However, this experimental method is not yet clinically feasible due to the long total acquisition time necessary to get sufficient SNR in particular in kidney imaging. Moreover, the bi-exponential behaviour of the diffusion MR signal in kidneys was shown before [26] and should be investigated in terms of tensorial analysis in future studies. In a recent study, Lanzman et al. could already highlight the potential of DT imaging for non-invasive functional assessment of transplanted kidneys [29]. They could show significant differences in FA values of the medulla between allograft recipients with heavily impaired renal function and those with moderate or mild impairment in renal function. Comparing MK values of normal kidneys with those of patients with various renal diseases may help to evaluate the clinical significance of renal kurtosis values and the role of the renal DKI. For instance in renal cancer, DKI may provide additional diagnostic information. Recently, Raab et al. [9] applied DK imaging in glioma and could differentiate between tumour grades using MK maps. Although the exact underlying meaning of the kurtosis findings has not yet been explained entirely these findings support the potential of DKI to reveal additional information to pathological alterations of the renal tissue. Since DKI has been proven to be more sensitive to tissue microstructure in comparison to ADC and FA measures, DKI of the kidney might be useful in evaluating conditions involving the renal tissue such as renal transplants, renal conditions after chemotherapy, or drug-induced renal tumours. Our study has some limitations. First, navigator-triggering was not possible due to technical limitations so that a respiratory belt had to be used. In clinical routine, the application of the respiratory belttype sensor may be suboptimal. Furthermore, respiratory movements of the kidney are mainly in a cranio-caudal direction, and do not always coincide with the abdominal wall movements. An approach that might further help reduce motion artifacts would be
G. Pentang et al. / Magnetic Resonance Imaging 32 (2014) 413–420
419
In conclusion, DKI of human kidneys is feasible. The use of 4 signal averages seems adequate in order to obtain good image quality when 3 b-values of (0, 300, 600 s/mm 2) and 30 diffusion directions are used.
Acknowledgments We thankfully acknowledge E. Rädisch for assistance with data acquisition.
References
Fig. 6. Example of a diffusion MR signal attenuation of the renal cortex (S/S0) against the b-value, including the mathematical fitting of the two models [1] and [2] to illustrate data modelling. The asterisks (*) represent the measured signal intensities within ROIs in the renal cortex (see Fig. 1) averaged over the 8 signal averages and all the volunteers; “mono-exp” and “kurt” denote the mono-exponential and the kurtosis model. The graph clearly illustrates the errors associated with the assumption of Gaussian distribution of water diffusion as in the case of the monoexponential fit (r = 0.96) versus a non-Gaussian distribution assumption from DKI (r = 0.99).
the use of navigator-echo type respiratory triggered acquisitions. They could help monitoring diaphragmatic motion and therefore decrease misregistration [25]. Due to technical difficulties on the MR scanner, this triggered acquisition technique could not be combined in our respiratory triggered DWI protocol. Second, the study was conducted on young healthy volunteers who were able to perform normal regular breathing. The results might differ in older subjects, patients in pain and patients who are less cooperative, having difficulties following a respiratory triggered acquisition. Further the hydration status of the kidney was not controlled in our study. It is known, that renal diffusion properties vary with water load. The influence of water load on the renal diffusion kurtosis still has to be investigated [17]. However, although we did not control the water load, our measurements of the reproducibility showed stable results of MK and FA within the error bounds. Further studies including patients with renal diseases should be conducted.
[1] Basser PJ, Mattiello J, LeBihan D. MR diffusion tensor spectroscopy and imaging. Biophys J 1994;66:259–67. [2] Basser PJ, Pierpaoli C. Microstructural and physiological features of tissues elucidated by quantitative-diffusion-tensor MRI. J Magn Reson B 1996;111: 209–19. [3] Jensen JH, Helpern JA, Ramani A, Lu H, Kaczynski K. Diffusional kurtosis imaging: the quantification of non-Gaussian water diffusion by means of magnetic resonance imaging. Magn Reson Med 2005;53:1432–40. [4] Jensen JH, Helpern JA. MRI quantification of non-Gaussian water diffusion by kurtosis analysis. NMR Biomed 2010;23:698–710. [5] Tabesh A, Jensen JH, Ardekani BA, Helpern JA. Estimation of tensors and tensorderived measures in diffusional kurtosis imaging. Magn Reson Med 2011;65: 823–36. [6] Tabesh A, Jensen JH, Ardekani BA, Helpern JA. Erratum: estimation of tensors and tensor-derived measures in diffusional kurtosis imaging. Magn Reson Med 2011;65:1507. [7] Cheung MM, Hui ES, Chan KC, Helpern JA, Qi L, Wu EX. Does diffusion kurtosis imaging lead to better neural tissue characterization? A rodent brain maturation study. NeuroImage 2009;45:386–92. [8] Hui ES, Cheung MM, Qi L, Wu EX. Towards better MR characterization of neural tissues using directional diffusion kurtosis analysis. NeuroImage 2008;42: 122–34. [9] Raab P, Hattingen E, Franz K, Zanella FE, Lanfermann H. Cerebral gliomas: diffusional kurtosis imaging analysis of microstructural differences. Radiology 2010;254:876–81. [10] Falangola MF, Jensen JH, Babb JS, Hu C, Castellanos FX, Di Martino A, et al. Agerelated non-Gaussian diffusion patterns in the prefrontal brain. J Magn Reson Imaging 2008;28:1345–50. [11] Qayyum A. Diffusion-weighted imaging in the abdomen and pelvis: concepts and applications. Radiographics 2009;29:1797–810. [12] Quentin M, Blondin D, Klasen J, Lanzman RS, Miese F-R, Arsov C, et al. Comparison of different mathematical models of diffusion-weighted prostate MR imaging. Magn Reson Imaging 2012;30:1468–74. [13] Yoshikawa T, Kawamitsu H, Mitchell DG, Ohno Y, Ku Y, Seo Y, et al. ADC measurement of abdominal organs and lesions using parallel imaging technique. AJR Am J Roentgenol 2006;187:1521–30. [14] Stejskal EO, Tanner JE. Spin diffusion measurements: spin echoes in the presence of a time-dependent field gradient. J Chem Phys 1965;42:288. [15] Denis Le Bihan, Mangin J-F, Poupon C, Clark CA, Pappata S, Molko N, et al. Diffusion Tensor Imaging: Concepts and Applications n.d. 2001. [16] Lagana M, Rovaris M, Ceccarelli A, Venturelli C, Marini S, Baselli G. DTI parameter optimisation for acquisition at 1.5 T: SNR analysis and clinical application. Comput Intell Neurosci 2010;2010.
Fig. 7. Bland-Altman analysis of the difference between the repeated measurements of the two observers.
420
G. Pentang et al. / Magnetic Resonance Imaging 32 (2014) 413–420
[17] Notohamiprodjo M, Dietrich O, Horger W, Horng A, Helck AD, Herrmann KA, et al. Diffusion tensor imaging (DTI) of the kidney at 3 Tesla – feasibility, protocol evaluation and comparison to 1.5 Tesla. Invest Radiol 2010;45:245–54. [18] Kido A, Kataoka M, Yamamoto A, Nakamoto Y, Umeoka S, Koyama T, et al. Diffusion tensor MRI of the kidney at 3.0 and 1.5 Tesla. Acta Radiol Stockh Swed 1987 2010;51:1059–63. [19] Gurses B, Kilickesmez O, Tasdelen N, Firat Z, Gurmen N. Diffusion tensor imaging of the kidney at 3 tesla: normative values and repeatability of measurements in healthy volunteers. Diagn Interv Radiol 2011;17(4):317–22. [20] Rosenkrantz AB, Sigmund EE, Johnson G, Babb JS, Mussi TC, Melamed J, et al. Prostate cancer: feasibility and preliminary experience of a diffusional kurtosis model for detection and assessment of aggressiveness of peripheral zone cancer. Radiology 2012;264:126–35. [21] Thoeny HC, Keyzer FD, Oyen RH, Peeters RR. Diffusion-weighted MR imaging of kidneys in healthy volunteers and patients with parenchymal diseases: initial Experience1. Radiology 2005;235:911–7. [22] Ries M, Jones RA, Basseau F, Moonen CTW, Grenier N. Diffusion tensor MRI of the human kidney. J Magn Reson Imaging 2001;14:42–9. [23] Müller MF, Prasad PV, Bimmler D, Kaiser A, Edelman RR. Functional imaging of the kidney by means of measurement of the apparent diffusion coefficient. Radiology 1994;193:711–5.
[24] Heusch P, Wittsack H-J, Kröpil P, Blondin D, Quentin M, Klasen J, et al. Impact of blood flow on diffusion coefficients of the human kidney: a time-resolved ECG-triggered diffusion-tensor imaging (DTI) study at 3 T. J Magn Reson Imaging 2013;37:233–6. [25] Kataoka M, Kido A, Yamamoto A, Nakamoto Y, Koyama T, Isoda H, et al. Diffusion tensor imaging of kidneys with respiratory triggering: optimization of parameters to demonstrate anisotropic structures on fraction anisotropy maps. J Magn Reson Imaging 2009;29:736–44. [26] Jansen JFA, Stambuk HE, Koutcher JA, Shukla-Dave A. Non-Gaussian analysis of diffusion-weighted MR imaging in head and neck squamous cell carcinoma: a feasibility study. Am J Neuroradiol 2010;31:741–8. [27] Wittsack H-J, Lanzman RS, Mathys C, Janssen H, Mödder U, Blondin D. Statistical evaluation of diffusion-weighted imaging of the human kidney. Magn Reson Med 2010;64:616–22. [28] Grinberg F, Farrher E, Kaffanke J, Oros-Peusquens A-M, Shah NJ. Non-Gaussian diffusion in human brain tissue at high b-factors as examined by a combined diffusion kurtosis and biexponential diffusion tensor analysis. NeuroImage 2011;57:1087–102. [29] Lanzman RS, Ljimani A, Pentang G, Zgoura P, Zenginli H, Kröpil P, et al. Kidney transplant: functional assessment with diffusion-tensor MR imaging at 3 T. Radiology 2013;266:218–25.