IMAGING METHODOLOGY Notes

Magnetic Resonance in Medicine 73:1526–1532 (2015)

Combined Intravoxel Incoherent Motion and Diffusion Tensor Imaging of Renal Diffusion and Flow Anisotropy Mike Notohamiprodjo,1,2,3* Hersh Chandarana,1 Artem Mikheev,1 Henry Rusinek,1 John Grinstead,4 Thorsten Feiweier,5 Jos e G. Raya,1 Vivian S. Lee,1,6 1 and Eric E. Sigmund INTRODUCTION

Purpose: We used a combined intravoxel incoherent motion– diffusion tensor imaging (IVIM-DTI) methodology to distinguish structural from flow effects on renal diffusion anisotropy. Methods: Eight volunteers were examined with IVIM-DTI at 3T with 20 diffusion directions and 10 b-values. Mean diffusivity (MD) and fractional anisotropy (FA) from DTI analysis were calculated for low (b  200 s=mm2), high (b > 200 s=mm2), and full b-value ranges. IVIM-parameters perfusion-fraction fP, pseudo-diffusivity Dp, and tissue-diffusivity Dt were first calculated independently on a voxelwise basis for all directions. After estimating a fixed isotropic fp from these data, global anisotropies of Dt and Dp in the cortex and medulla were determined in a constrained cylindrical description and visualized using polar plots and cosine scatterplots. Results: For all b-value ranges, medullary FA was significantly higher than that of the cortex. The corticomedullary difference was smaller for the high b-value range. Significantly higher fp and Dt were determined for the cortex and showed a significantly higher directional variance in the medulla. Polar plot analysis displayed nearly isotropic Dp and Dt in the cortex and anisotropy in the medulla. Conclusion: Both flow and microstructure apparently contribute to the medullary diffusion anisotropy. The described novel method may be useful in separating decreased tubular flow from irreversible structural tubular damage, for example, in diabetic nephropathy or during allograft rejection. Magn Reson C 2014 Wiley Periodicals, Inc. Med 73:1526–1532, 2015. V Key words: diffusion; IVIM; DTI; kidney; medulla; anisotropy; microcirculation

1 Center for Biomedical Imaging, Department of Radiology, NYU Langone Medical Center, New York, New York, USA. 2 Department of Clinical Radiology, University Hospitals Munich, Munich, Germany. 3 Department of Radiology, University Hospital Tuebingen, Tuebingen, Germany. 4 Siemens AG, Healthcare Sector, Portland, Oregon, USA. 5 Siemens AG, Healthcare Sector, Erlangen, Germany. 6 University of Utah School of Medicine, Salt Lake City, Utah, USA.

*Correspondence to: Mike Notohamiprodjo, M.D., Institute of Clinical Radiology, Ludwig-Maximilians-University Hospital Munich, Germany, Marchioninistrasse 15, 81377 Munich, Germany. E-mail: mike.notohamiprodjo@ med.uni-muenchen.de Additional Supporting Information may be found in the online version of this article. Received 22 July 2013; revised 19 February 2014; accepted 17 March 2014 DOI 10.1002/mrm.25245 Published online 21 April 2014 in Wiley Online Library (wileyonlinelibrary. com). C 2014 Wiley Periodicals, Inc. V

Diffusion weighted imaging (DWI) characterizes water motion on a molecular level and provides information on renal microstructure and function (1,2). The apparent diffusion coefficient (ADC) serves as a convenient measure and shows alteration in renal pathology such as renal artery stenosis (3), renal failure (4), ureteral obstruction (5), and transplant rejection (6). However, the ADC provides a mix of information on microcirculation and tissue parenchymal structure and cellularity (7), and other approaches are warranted to resolve this ambiguity. Solving this ambiguity may be useful to distinguish reduced medullary flow from irreversible tissue damage, for example, in the diagnosis of allograft rejection or diabetic nephropathy. One variant of DWI analysis, intravoxel incoherent motion (IVIM) (7), distinguishes pseudodiffusion (tubular=vascular flow) from passive structural diffusion by collecting data over a range of diffusion weightings (i.e., b-values). The data is analyzed by bi-exponential fitting of the signal decay to account for the fast pseudodiffusion component (typically relevant for b < 200 s=mm2) and the slow tissue diffusion component. IVIM parameters have shown sensitivity to renal allograft rejection (8,9) and vascularity=cellularity of renal masses (10,11). Another DWI technique, diffusion tensor imaging (DTI) (12), can measure the anisotropy imposed on the water diffusion by the tissue microstructure through analysis of multiple measurements with different diffusion-sensitizing directions. DTI demonstrates clear anisotropy in the renal medulla (13,14). Decreasing diffusion anisotropy is observed in diabetic nephropathy (15), allograft dysfunction (16,17), ischemia-reperfusion damage (18), and chronic parenchymal disease (19). However, the biophysical underpinning of this anisotropy remains unproven, particularly regarding the roles of a) structural restrictions of tubules and collecting ducts (20) and b) active flow in oriented tubular or vascular structures (18). Several studies observed a dependency of fractional anisotropy (FA) of the applied b-values (21,22), which is consistent with a role of flow effects in the medullary diffusion anisotropy. A recent study by Sigmund et al. described alteration of medullary FA anisotropy with flow challenge, possibly caused by changes in tubular diameter, but also potentially involving vascular flow, water reabsorption, and intratubular flow (23). Each process could play an important role in DWI=DTI contrast in the kidney (23). Resolving the ambiguity of diffusion anisotropy requires a more comprehensive acquisition and analysis approach. In

1526

IVIM-DTI of the Kidney

the present Note, we used a combined IVIM-DTI methodology to distinguish structural effects from flow effects on renal tissue anisotropy. We hypothesize that both components contribute to the renal diffusion anisotropy. Separating these effects may be useful to distinguish reduced medullary flow from irreversible tissue damage, for example, in the diagnosis of allograft rejection or diabetic nephropathy.

1527

The acquired DWI-signal decay curves were calculated for each acquired direction. IVIM-parameters perfusion fraction fp, pseudo-diffusivity Dp, and tissue diffusivity Dt were calculated with a voxelwise biexponential analysis (7). The model taken for the magnetization M has four parameters: total magnetization M0, perfusion fraction fp, pseudo-diffusivity Dp, and tissue diffusivity Dt :

METHODS Study Population This Health Insurance Portability and Accountability Actcompliant study was approved by the local institutional review board. After giving written informed consent, eight healthy volunteers (16 kidneys) (3 male, 5 female; age 30.6 6 5.2; range 24–36 years) were included in this study. The volunteers had no history of renal disease. To standardize the hydration status, consumption of caffeine or tea was prohibited 3 hours before examination, and the volunteers were orally hydrated with 150 ml of water 15 minutes prior to the acquisition. MR Imaging The volunteers were examined using a clinical wide-bore 3T-scanner (Verio, Siemens Healthcare, Erlangen, Germany). A six-element body array matrix coil and 12 elements of the integrated spine coil were used for signal acquisition. A coronal single-shot half-Fourier acquisition single-shot turbo spin-echo (HASTE) sequence (repetition time [TR] 1030 ms; echo time [TE] 99 ms; matrix 384  384; 20 slices; voxel size 1.1  1.1  5 mm3) was acquired for morphology and planning of the IVIM-DTI sequence. IVIM-DTI was acquired during free-breathing using a prototype twice-refocused spin echo echo-planar imaging (EPI) sequence with reversed slice gradient polarity fat suppression and nonlinear phase correction (24): (TR ¼ 2600 ms; TE ¼ 79 ms; 2 averages; b-values 0, 10, 30, 50, 80, 120, 200, 400, 600, 800 s=mm2; all b-values > 0 s=mm2 were acquired along 20 diffusion directions; slice thickness ¼ 6 mm; field of view ¼ 420  420 mm2; matrix ¼ 156  192 interpolated to 312  384; voxel size ¼ 2.1  2.1  6 mm3 (interpolated to 1  1  6 mm3); number of slices ¼ 5; bandwidth ¼ 1860 hz=pixel; phase-encoding direction left to right; parallel imaging GRAPPA factor R ¼ 2). Acquisition time for the IVIM-DTI sequence was 15 minutes. Data Analysis To mitigate respiratory motion artifacts, retrospective twodimensional affine registration of EPI images was performed separately for the right and left kidney with in-house software (FireVoxel; CAI2R, New York University, NY) (25). IVIM-DTI analysis of the renal cortex and medulla was performed using custom code (IgorPro 6; Wavemetrics, Inc., Portland, OR). To improve signal-to-noise ratio, a median (nonlinear edge preserving) filter of size 3  3 voxels was applied to each DW image before processing. A single investigator (M.N.) defined regions of interest (ROIs) on a central b0-slice on the cortex (n ¼ 1) and the medulla (n ¼ 3). The same ROIs were used for all variants of analysis performed. IVIM analysis First, IVIM-metrics were derived for each voxel and diffusion-sensitizing direction independently as follows.

M ¼ M0 ðfp expðbDp Þ þ ð1  fp ÞexpðbDt ÞÞ

[1]

A segmented IVIM analysis was performed, as described previously (10,26–28), to ensure a more robust analysis compared to an unconstrained fit, albeit at the expense of some accuracy due to the assumptions involved in the separate measurement of Dt, as follows. When the bvalue is significantly greater than 1=Dp (e.g., for Dp ¼10 mm2=ms; 100 s=mm2), the pseudodiffusion term is small, so that Eq. [1] can be simplified: Mhigh ¼ M0 ð1  fp ÞexpðbDt Þ

[2]

Dt was determined from a monoexponential fit of the asymptotic high b-values range (b > 200 s=mm2). Its zero intercept M0 (1fp) 5 Mint is used along with the unweighted (b ¼ 0) signal M0 to determine fP. fp ¼

M0  Mint M0

[3]

Dp-values were calculated from a biexponential fit with constrained Dt and fp according to Eq. [1]. Parametric maps of the mean Dt, fp, and Dp over all directions were generated.

DTI analysis Next, standard DTI analysis was performed with DWI from different ranges of b-values to gauge the effect of flow on apparent anisotropy. Specifically, ADC values along each direction were calculated from the full range of diffusion-weighted images (b ¼ 0–800 s=mm2) and for images without (b > 200 s=mm2) and with some (b  200 s=mm2) pseudodiffusion contribution:   M ln ¼ b  ADC [4] M0 The diffusion tensor was calculated from these derived ADC-values by fitting the data to a 3  3 symmetric tensor model (12,29). The properties of the tensor were v 2; ! v 3 ) and its characterized by its eigenvectors (! v 1; ! eigenvalues (l1, l2, l3)—the orientation and magnitude of diffusion along the principal axes. The primary eigenvalue l1 (axial diffusivity) is the largest and leastrestricted component—and in the medulla presumably includes motion along the tubular axis, for example, tubular flow or vascular flow along the vasa recta (18,20). Maps of mean diffusivity (MD), fractional anisotropy (FA), and primary eigenvector were generated as per standard calculations (12).

1528

Notohamiprodjo et al.

Table 1 Average Kidney IVIM Parameters Derived From 20 Directions

Cortex

Medulla

2.44 6 0.12 22.65 6 10.63 26.6 6 6.1

2.18 6 0.17a 28.20 6 8.13 14.1 6 4.5b

Dt and Dp in 103 mm2=s, fp in %. a Indicates statistical significant corticomedullary difference (P < 0.05). b P < 0.01.

1 MD  ðl1 þ l2 þ l3 Þ 3 vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  u  2 u3 ðl1  MDÞ þ ðl2  MDÞ2 þ ðl3  MDÞ2 t FA ¼ 2ðl21 þ l22 þ l23 Þ

[5] [6]

in the literature [e.g., Ref. (23), in which 2*(l2–l3) = (l2þ l3) ¼ 0.15, while 2*(l1– l2)= (l1þ l2) ¼ 0.32]. Finally, histograms and scatterplots of the Dt and Dp values versus cos(u) were generated. In an equivalent procedure confirming the anisotropy derived from the polar plots, a quadratic fit Di ¼ Di:axial þ ðDi:axial  Di;radial Þcos2 u was applied to the cos(u) scatterplots for both Dt and Dp to derive their global axial and radial values. Statistical Analysis We used the Wilcoxon signed rank test to determine significant differences between the cortex and medulla of the parameters described above. Statistical significance was defined by P < 0.05. Statistical analysis was performed with SPSS 15.0 (SPSS, Inc., Chicago, IL). RESULTS

IVIM-DTI anisotropy analysis and visualization Finally, informed by the separate IVIM and DTI analyses, a cylindrical two-compartment description was adopted as follows to quantify the global anisotropies of the diffusion and perfusion components. In this approach, the perfusion fraction fp describing the effective spin density (weighted by relaxation, coil sensitivity, etc.) of the flow compartment is fixed as an isotropic scalar (see below). The anisotropy of the flow part of the IVIM signal is represented by the direction dependence of the pseudodiffusivity Dp. The joint IVIM-DTI analysis was conducted as follows. First, each diffusion direction was fitted with the segmented IVIM algorithm as described earlier, providing the respective fp, Dt, and Dp values. Next, a diffusion tensor analysis was performed on the tissue diffusivity from the (Dt) values, and its principal eigenvector direction was determined. The average value fp,ave over all measured directions was adopted as the “true” scalar fp,s. Finally, a second IVIM fit was performed to each gradient direction constraining the same scalar fp,s for all directions and the previously measured Dt for each direction, to extract the directional Dp values. Next, the Dt and Dp values from all ROI (cortex or medulla analyzed separately) voxels in 20 directions (total > 1,000 data points=subject) were plotted as a function of the polar angle u ¼ arccosð^ g  v^ 1 Þ between the diffusion gradient direction and the tensor direction (i.e., a “peanut plot”) (30,31) approximating the Dp dependence as a cylindrical tensor along the Dt principal direction. Because each voxel’s data points are displayed relative to its v1 vector, all ROI voxels can be plotted=analyzed together in the same polar plot. For visualization of the entire polar plane and consistent with the cylindrical description, data points at each angle u were reflected around the polar axis (u!u,180 þ u,180u). The mean and standard error (standard deviation=(number of points=4)0.5, accounting for the reflected u points) within angular bins (covering 9 ) of all ROI voxels displayed in a polar plot were calculated. Global FA values reflecting Dt and Dp anisotropy were estimated from the 0 (axial, along v1) and 90 (radial, perpendicular to v1) values of the averaged angular distribution, using Eq. [6] with assumed cylindrical symmetry (l2l3) as supported

All examinations were diagnostic, no considerable artifacts due to misregistration or distortion were observed. No kidney was excluded from the analysis. Table 1 summarizes data from the separate IVIM analyses for all eight volunteers in this study. The data of the DTI is analysis is provided in the Supporting Table S1. Dt, fp, MD, and FA showed significant corticomedullary contrast (Fig. 1). Medullary FA was significantly higher (P < 0.01) than that of the cortex for the full b-value set, the perfusion-weighted low b-value set, and the tissue-weighted high b-value regime. However the corticomedullary difference was smaller for the high b-value range (Table 1 and Fig. 1). MD values of both the medulla and cortex were significantly higher for the low b-value range than for other ranges (P < 0.01). Significantly higher directionally averaged fp (P < 0.01), Dt, and MD (P < 0.05) values were found for the cortex than for the medulla (Fig. 2). Average Dp showed a high intraindividual and interindividual variance and did not show significant corticomedullary contrast. The polar plot analysis of the cylindrical twocompartment description depicts nearly isotropic DP and Dt in the cortex. In the medulla, a distinct peanut shape of the polar plot was observed, suggesting anisotropy for both Dt and Dp parameters in the medulla (Fig. 3). Importantly, the flow component (Dp) shows anisotropy along the same direction as the structural component (Dt) because the polar angles u were defined by the latter’s tensor orientation. FA derived from the 0 (axial, along v1) and 90 (radial, perpendicular to v1) values of the averaged angular distribution showed a significant corticomedullary difference for both Dt and Dp. Equivalently, both Dt and Dp cos(u) scatterplots show strong quadratic behavior for the medulla indicating anisotropy, in contrast to the much weaker angular dependence of the data derived from the cortex (Table 2 and Supporting Fig. S1). Axial and radial diffusivities derived from quadratic fits were equivalent to those measured from polar plots within 3% to 5% in each case, as expected for the same underlying data presented in two ways. DISCUSSION Using a combined protocol including 10 b-values and 20 different directions with retrospective motion correction,

IVIM-DTI of the Kidney

1529

FIG. 1. DTI as a function of b-value. 24-year-old female volunteer: Fractional anisotropy (FA) of the medulla was significantly higher than that of the cortex for all three b-value ranges (a, b, c). However, the corticomedullary difference of the perfusion-weighted low b-value range (0  b  200 s=mm2, (a) was higher than for the tissue weighted high b-value range (200

Combined intravoxel incoherent motion and diffusion tensor imaging of renal diffusion and flow anisotropy.

We used a combined intravoxel incoherent motion-diffusion tensor imaging (IVIM-DTI) methodology to distinguish structural from flow effects on renal d...
438KB Sizes 0 Downloads 3 Views