Original Investigation

The Study of Clear Cell Renal Cell

Carcinoma with MR Diffusion
Kurtosis Tensor Imaging and Its
Histopathologic Correlation
Guangyu Wu, MD, Zizhou Zhao, BM, Qiuying Yao, BM, Wen Kong, MD, Jianrong Xu, PhD,
Jin Zhang, PhD, Guiqin Liu, MD, Yongming Dai, PhD

Rationale and Objectives: The objective of this study was to compare the performance of diffusion kurtosis tensor imaging and diffusion-
weighted imaging in the characterization of clear cell renal cell carcinoma (ccRCC) and their correlations with tumor histopathology.
Materials and Methods: Ninety-one patients diagnosed with ccRCC who underwent diffusion kurtosis tensor imaging were included
in this study. Fractional anisotropy, mean diffusivity, radial diffusivity, axial diffusivity, mean kurtosis (MK), radial kurtosis (Krad), and
axial kurtosis (Kax) data were produced. A nuclear grade of 1–4 (G1–4) was assigned for each case based on the Fuhrman grading
system, whereas tumor histopathology was characterized by the nuclear-to-cytoplasm ratio, the cell nuclei count, and the cell volume
fraction.
Results: All of the metric values except for Kax and fractional anisotropy could be used to discriminate G1 vs G3, G1 vs G4, G2 vs
G3, and G2 vs G4, whereas MK and Kax could be used to discriminate G3 vs G4 (P < 0.05). Moreover, the MK and Krad values ex-
hibited better performance in differentiating G2 from G3 (P < 0.04 compared to the other metrics). The nuclear-to-cytoplasm ratio was
positively correlated with the MK, Krad, and Kax values (P < 0.001) and negatively correlated with the mean diffusivity, radial diffusivity,
and axial diffusivity values (P < 0.001), whereas the cell volume fraction and the cell nuclei count did not correlate with any metric examined.
Conclusion: The kurtosis metrics were superior to the diffusion metrics in grading ccRCC.
Key Words: Clear cell Renal cell carcinoma; diffusion kurtosis tensor imaging; histopathologic correlation.
© 2018 The Association of University Radiologists. Published by Elsevier Inc. All rights reserved.

INTRODUCTION treatment of RCC (2,3). Accurately identifying the tumor grade

in ccRCC by magnetic resonance imaging (MRI) would

R
enal cell carcinoma (RCC) is the most lethal of the
provide a potential advantage in treatment decisions in the
urologic malignancies and comprises various sub-
future, as well as a useful modality for related research fields.
types, among which clear cell renal cell carcinoma
MRI is a noninvasive technique used to diagnose renal tumors.
(ccRCC) accounts for the majority (70%). Different ccRCC
In particular, diffusion-weighted imaging (DWI) can be used
grades exhibit diverse biological behaviors and variable clin-
to further characterize biological tissue structures by measur-
ical outcomes (1). Meanwhile, minimally invasive techniques,
ing the diffusion of water molecules (4–6). Previous studies
such as percutaneous radiofrequency, cryoablation, micro-
on renal tumors have shown that the apparent diffusion co-
wave therapy, and high-intensity focused ultrasound ablation,
efficient (ADC) value derived from DWI signals may provide
have been suggested as feasible alternatives to the surgical
a quantitative method for differentiating benign from malig-
nant tumors (7) and identifying the histopathologic degree of
Acad Radiol 2018; 25:430–438
ccRCC (8). However, the performance of DWI is limited
From the Department of Radiology, Renji Hospital, School of Medicine,
Shanghai Jiao Tong University, No. 1630, Dongfang Road, Pudong, Shanghai because of the confounding overlap of broad ADC ranges
200120 (G.W., Z.Z., Q.Y., J.X., G.L.); Department of Urology, Renji Hospital, (9,10).
School of Medicine, Shanghai Jiao Tong University, Shanghai, China (W.K.,
J.Z.); MR Collaboration, United Imaging Healthcare, Shanghai, China (Y.D.).
Conventional DWI assumes that water diffusion has a Gauss-
Received August 27, 2017; revised October 13, 2017; accepted October 20, ian distribution. However, because of the microstructural
2017. Dr. Guangyu Wu and Dr. Zizhou Zhao contributed equally to this article. complexity of tissues and cells, including cell membranes, in-
Funding Source: This work is supported by the National Natural Science
Foundation of China (81601487/81371622). Address correspondence to: tracellular organelles, and water compartments, the diffusion
J.X. e-mail: renjixujr@163.com; G.L. e-mail: sdzclgq@126.com; J.Z. e-mail: of water molecules tends to deviate from a Gaussian distri-
med-zhangjin@vip.sina.com
bution (11), thereby limiting the effectiveness of conventional
© 2018 The Association of University Radiologists. Published by Elsevier Inc.
All rights reserved.
DWI. Kurtosis is a dimensionless statistical metric for quan-
https://doi.org/10.1016/j.acra.2017.10.016 tifying the non-Gaussianity of an arbitrary probability

430
Downloaded for Siti Fadhilah (fadhilahhazhiyah@gmail.com) at Universitas Muslim Indonesia from ClinicalKey.com by Elsevier on August 30, 2018.
For personal use only. No other uses without permission. Copyright ©2018. Elsevier Inc. All rights reserved.
Academic Radiology, Vol 25, No 4, April 2018 DKT IMAGING AND DWI IN CCRCC

distribution (12), and diffusion kurtosis tensor (DKT) imaging
is an extension of diffusion tensor imaging (DTI) that quan-
tifies non-Gaussian water diffusion by acquiring data for at
least two nonzero diffusion gradient factors (b values) in at
least 15 independent directions. The kurtosis metrics (ie, the
mean kurtosis [MK], axial kurtosis [Kax], and radial kurtosis
[Krad]) and the conventional diffusion metrics (ie, the mean
diffusivity [MD], axial diffusivity [Dax], radial diffusivity [Drad],
and fractional anisotropy [FA]) can be derived from DKT
imaging data simultaneously. The Kax and the Krad are par-
allel and perpendicular to the main direction of diffusion,
respectively, whereas the MK is the average kurtosis over all
diffusion directions (13). Accordingly, DKT imaging may be
more suitable than DWI and DTI for evaluating microstruc-
tural changes in tissues. DKT imaging has offered numerous
diagnostic possibilities in the liver, kidneys, and prostate gland
(14–17). Although a recently published study indicated that
it is feasible to use diffusion kurtosis imaging (DKI) to assess
the ccRCC grade (18), that particular study did not use a
tensor-based method, which precludes the evaluation of the
Krad and the Kax. The value of DKT imaging in the assess-
ment of ccRCC has not yet been comprehensively reported.
The purpose of our study was to explore the value of the dif-
fusion and kurtosis metrics in the characterization of ccRCC
with different grades and to correlate them with the cytoar-
chitectural differences in tumor grades.
images with obvious artifacts, (2) antitumor therapy and a biopsy
performed before MRI, (3) unavailability of dynamic contrast-
enhanced (DCE) MRI data, and (4) solid components that
were difficult to characterize (eg, a cystic lesion with mural
linear enhancement in DCE MRI). The present study was
approved by our institutional review board, and the require-
ment for written informed consent from patients was waived.
MRI Protocol
MRI was performed on a 1.5-T magnetic resonance (MR)
scanner (uMR 550; United Imaging Healthcare, Shanghai,
China) with a commercialized 12-channel body phased-
array coil. Conventional sequences included axial T1-
weighted in- and out-phase imaging, axial T2-weighted
imaging, and axial and coronal fat saturation fast spin echo
T2-weighted imaging. DKT was conducted based on a con-
ventional DTI sequence with three b values in 32 directions.
Multiphase DCE MR images were acquired at 20, 45, 120,
and 180 seconds after intravenous administration of
gadopentetate dimeglumine (Magnevist; Schering, Berlin,
Germany; total dosage = 0.2 mL/kg, flow velocity = 2.5 mL/s)
using axial three-dimensional fat saturation T1-weighted spoiled
gradient echo. Detailed MRI protocols are listed in Table 1.
All patients underwent surgery within 1 week after the MR
scan.

MATERIALS AND METHODS

Patients
From June 2015 to June 2016, 135 consecutive patients were
suspected to have RCC and had undergone MRI, includ-
ing DKT imaging; among these patients, those diagnosed with
ccRCC were included in our study and retrospectively evalu-
ated by two experienced radiologists (with more than 5 years
of experience) who were blinded to the pathologic results.
The exclusion criteria were as follows: (1) poor-quality DTI
Imaging Analysis
DKI data were processed using Diffusional Kurtosis Estima-
tor (version 2.5.1; Medical University of South Carolina,
Charleston, SC, USA), and images were processed using ImageJ
software (version 1.49b, National Institutes of Health, Bethesda,
Maryland, USA). Metric maps were calculated for the MK,
Kax, Krad, MD, Dax, Drad, and FA (19).
The region of interest (ROI) was placed within the solid
tumor area on the DCE images. The ROI was either circu-

TABLE 1. Protocols of Magnetic Resonance Sequences

T1-Weighted In-phase or T2-Weighted T2-Weighted Diffusion Tensor Contrast-enhanced

Sequence Opposed-phase Image Imaging Imaging Imaging Imaging
Scan plane Axal Axial Coronal Axial Axial
TR/TE 180/2.1, 4.4 2308/79 1500/85 3300/57 6/2
FOV (cm) 38~42 38~42 42 38~42 38~42
Thickness (mm) 5 5 4 5 2.5
Spacing (mm) 1 1 0.4 1 0
Matrix 512 × 512 512 × 512 512 × 512 128 × 128 512 × 512
NEX 1 3 1 2 1
Bandwidth 62.5 44 83.3 125 125
Fat saturation None Yes None Yes Yes
Flip angle 80 90 90 90 15
Other b Values: 0, 300, 600
Number of directions: 32

FOV, field of view; NEX, number of excitations; TE, echo time; TR, repetition time.

431
Downloaded for Siti Fadhilah (fadhilahhazhiyah@gmail.com) at Universitas Muslim Indonesia from ClinicalKey.com by Elsevier on August 30, 2018.
For personal use only. No other uses without permission. Copyright ©2018. Elsevier Inc. All rights reserved.
WU ET AL Academic Radiology, Vol 25, No 4, April 2018

Figure 1. A 47-year-old man with G1 clear cell renal cell carcinoma in the left kidney. (a) T2-Weighted image, (b) non–contrast-enhanced
T1-weighted image, and (c) contrast-enhanced T1-weighted image acquired at 45 seconds. Mapping of the (d) mean diffusivity, (e) mean
kurtosis, (f) radial diffusivity, (g) axial diffusivity, (h) radial kurtosis, (i) axial kurtosis, and (j) fractional anisotropy. Regions of interest were
drawn in the tumor areas and in the normal renal cortical and medullary areas in the contralateral kidney.

lar or elliptical, with an area of 30–120 mm2. ROI placement
was avoided in necrotic and cystic regions identified with con-
ventional MRI sequences (T2-weighted imaging and DCE
MRI). For comparison, an ROI with a similar size was also
placed in a healthy portion of the renal cortex and medulla
in the contralateral kidney and in the corresponding upper
or lower pole if the tumor bulged remarkably outside of the
contour of the affected kidney. None of the ROIs were po-
sitioned on the boundary between the cancerous and normal
parenchyma to avoid partial volume averaging. The values
of the diffusion and kurtosis metrics within an ROI were re-
corded (Fig 1). Lesion sizes were defined as the longest
dimension on T2-weighted images. All ROI-related mea-
surements were performed twice with an interval of 3 days
to assess intraobserver agreement by two radiologists (with more
than 5 years of experience in abdominal MRI) who were
blinded to the pathologic results.
Histologic Results
After surgery, the histologic specimens were sectioned and
stained with hematoxylin and eosin for subsequent exami-
nation by light microscopy. A single uropathologist (with 10
years of experience in uropathology) who was blinded to the
previous MRI findings reviewed all histologic slides (400×)
using an optical microscope (Nikon Eclipse E 600; Nikon,
Osaka, Japan). Images were then digitally captured, and a
nuclear grade of 1–4 (G1–4) was assigned for each case based
on the Fuhrman grading system (17). Histologic analysis was
performed without prior knowledge of the MRI results by
an experienced pathologist (8 years of experience in histol-
ogy). Tumor histopathology was analyzed based on the digital
images of the stained histologic slides of each renal lesion using
Image-Pro Plus software (version 6.0; Media Cybernetics, Inc.,
Washington, DC). Adaptive histogram equalization and a color
intensity threshold were used to segment darker-stained cell
nuclei (Fig 2) to automatically obtain the nuclear-to-
cytoplasm (N/C) ratio, the cell nuclei count, and the cell
volume fraction (CVF) in 10 high-magnification (400×) fields.
Statistical Analysis
MedCalc Statistical Software (version 15.6.1; MedCalc Soft-
ware bvba, Ostend, Belgium) was used for all statistical analyses.
All metrics were expressed as the mean ± standard devia-
tion. The intra- and interobserver agreement of all metrics
was evaluated using the intraclass correlation coefficient (ICC)
(20). An ICC greater than 0.75 was considered to represent
good agreement (21). A receiver operating characteristic curve
analysis was used to assess the ability of the diffusion and kur-
tosis metrics to distinguish normal renal parenchyma from
ccRCC and the different grades of ccRCC. A simple linear
regression analysis was performed to calculate the Pearson cor-
relation coefficient between all metrics and changes in tumor
histopathology in the individual lesions. P < 0.05 was con-
sidered to represent statistical significance.
RESULTS
Twenty-nine patients were proven to have a renal tumor other
than ccRCC (angiomyolipoma: n = 8, oncocytoma: n = 2, pap-
illary RCC: n = 10, and chromophobe RCC: n = 9) and were
thus excluded from the study. A total of 106 patients were
diagnosed with ccRCC. Patients with poor-quality DTI images
with obvious artifacts (n = 5), patients who underwent anti-
tumor therapy and who were biopsied before undergoing MRI

432
Downloaded for Siti Fadhilah (fadhilahhazhiyah@gmail.com) at Universitas Muslim Indonesia from ClinicalKey.com by Elsevier on August 30, 2018.
For personal use only. No other uses without permission. Copyright ©2018. Elsevier Inc. All rights reserved.
Academic Radiology, Vol 25, No 4, April 2018
(n = 3), patients who did not have DCE MRI data (n = 3),
and patients with solid components that were difficult to char-
acterize (n = 4) were excluded. Thus, the final cohort comprised
91 patients (63 men [mean age, 60 years; age range, 26–85
years] and 28 women [mean age, 61 years; age range, 24–
85 years]) with an average age of 60.7 years (age range, 24–
85 years) and an average tumor diameter of 24 mm (diameter
range, 8–93 mm). The pathologic results of all patients in our
study cohort were obtained and included grades 1 (n = 21),
2 (n = 43), 3 (n = 19), and 4 (n = 8).
Table 2 shows the ccRCC diffusion and kurtosis metrics
and the inter-reader agreement for each parameter. The intra-
and interobserver ICCs of all metrics calculated on the basis
of the reader’s two measurements were good, ranging from
0.86 to 0.983. Therefore, all outcomes were based on the first
measurements of the reader.
The results of the differential normal renal parenchyma and
different ccRCC grades are shown in Table 3. For the normal
renal parenchyma, one patient was excluded because of con-
tralateral renal atrophy. The MK, Krad, Kax, and FA values
were significantly higher in the medulla than in the cortex,
whereas the MD, Drad, and Dax values were significantly
higher in the cortex than in the medulla. Upon comparison
of the ccRCC and normal renal parenchyma, the MD and
Dax values were significantly lower in the ccRCC tissue than
in the normal renal cortex and medulla; however, the MK,
Krad, and Kax values showed opposite results. The differ-
ence in the Drad between ccRCC and the normal medulla
was not prominent. In the discrimination of different ccRCC
grades, the MD, MK, and Krad could be used to differenti-
ate ccRCC in terms of G1 vs G3, G1 vs G4, G2 vs G3, and
G2 vs G4, whereas both the MK and Krad showed superior
performance in differentiating G2 vs G3 compared to the MD
Downloaded for Siti Fadhilah (fadhilahhazhiyah@gmail.com) at Universitas Muslim Indonesia from ClinicalKey.com by Elsevier on August 30, 2018.
For personal use only. No other uses without permission. Copyright ©2018. Elsevier Inc. All rights reserved.
DKT IMAGING AND DWI IN CCRCC
Figure 2. Tumor histopathology analy-
sis for cases of clear cell renal cell
carcinoma with different grades. Histo-
logic slides (400×) were stained with
hematoxylin and eosin, and tumor cell nuclei
were automatically segmented using adap-
tive histogram equalization and a color
intensity threshold: (a) grade 1, (b) grade
2, (c) grade 3, and (d) grade 4. (Color
version of figure is available online.)
(P = 0.039 and 0.017 vs MD; P = 0.042 and 0.026 vs Drad;
P < 0.001 vs the other metrics). It is also worthwhile to note
that the differences in the ability of MK and Kax to distin-
guish between G3 and G4 were also statistically significant
and superior to those of the other metrics (P = 0.010 and 0.007
vs MD; P < 0.001 vs the other metrics). For G1 vs G2, al-
though MK and Kax exhibited slight improvements over the
other metrics, the results did not reach a satisfactory diag-
nostic level. The corresponding receiver operating characteristic
plots are shown in Figure 3. Table 4 presents the changes in
tumor histopathology and the correlations between tumor his-
topathology and the diffusivity and kurtosis metrics. The N/C
ratio was positively correlated with the MK, Krad, and Kax
values and negatively correlated with the MD value, whereas
the CVF and the cell nuclei count were found to have no
obvious correlations with any DKT imaging or DTI metric
examined.
DISCUSSION
A previous study demonstrated the ability of the MK to detect
diffusion properties among phantoms with different degrees
of structural changes, which was not reflected by the MD,
indicating that the MK may be a more specific indicator of
tissue microstructural complexity (13). DKI has been pro-
posed to have potential for demonstrating the non-Gaussian
behavior of water diffusion in the evaluation of renal tumors
(15,16,18). However, because diffusion kurtosis is a tensorial
quantity, the previous study used a non–tensor-based method,
a proposed nonlinear least squares fitting procedure with DWI
data with a multi-b value, instead of a tensor-based method,
which raises several issues, namely, the precision and
accuracy in the metrics estimation. Moreover, the
433
WU ET AL Academic Radiology, Vol 25, No 4, April 2018

CNC, cell nuclei count; CVF, cell volume fraction; Dax, axial diffusivity; Drad, radial diffusivity; FA, fractional anisotropy; Kax, axial kurtosis; Krad, radial kurtosis; MD, mean diffusivity;
non–tensor-based method precludes the possibility of evalu-

Intraobserver Agreement
Interobserver and
ating the Krad and Kax values, which can provide information
of diffusion kurtosis in a particular direction in addition to
0.903/0.895
0.860/0.869
0.901/0.874
0.983/0.905
0.966/0.895
0.936/0.874
0.910/0.866
NA
NA
NA
the MK value. To our knowledge, the application of DKT
imaging in the present study is the first to evaluate all kur-
tosis metrics for the grading of ccRCC.
Generally, tumors with higher grades have worse progno-
ses. The previous investigation by Rosenkrantz et al. revealed
that the ADC values of high-grade ccRCC (ie, grades 3 and
4) were significantly lower than those of low-grade ccRCC
2.78 ± 0.37
0.89 ± 0.24
0.93 ± 0.36
0.84 ± 0.18
2.23 ± 0.35
3.87 ± 0.70
0.35 ± 0.09
(ie, grades 1 and 2) (22). A study by Yu et al. (10) analyzed
Medulla

the differences in ADC values between every pathologic grade

of ccRCC in detail, and the results showed that the ADC
NA
NA
NA
values decreased with increasing pathologic grades from 1 to
4 with significant differences between them, except between
grades 1 and 2 and between grades 3 and 4 (13). However,
3.55 ± 0.57
0.70 ± 0.14
0.67 ± 0.20
0.70 ± 0.15
3.02 ± 0.51
4.62 ± 0.79
0.28 ± 0.07
Cortex

in our study, the results showed that it is still challenging for

the diffusivity and kurtosis metrics to differentiate G1 from
NA
NA
NA

G2. Moreover, the value of diffusivity metrics was limited

for discriminating G3 vs G4, which was in agreement with
the results of a previous study (13). For the differentiation
388.53 ± 129.19

between G1 and G2, a potential explanation might be that

2.47 ± 0.72
1.17 ± 0.48
1.10 ± 0.46
1.22 ± 0.39
2.08 ± 0.70
3.35 ± 0.82
0.30 ± 0.19
0.37 ± 0.16

0.63 ± 0.18
TABLE 2. Diffusion and Kurtosis Metrics of Clear Cell Renal Cell Carcinoma and Normal Renal Parenchyma

the nuclear features are rather similar in these two situa-

All

tions, and slight differences in the size of the nuclei may not
lead to notable changes in tissue complexity, rendering it dif-
ficult for metrics to discriminate between the two grades.
Moreover, the MD values in tumors would be affected by
403.55 ± 115.80

several factors, and the potential information regarding dif-

1.58 ± 0.50
1.89 ± 0.42
1.73 ± 0.50
1.79 ± 0.31
1.16 ± 0.44
2.33 ± 0.67
0.30 ± 0.10
0.57 ± 0.10

0.67 ± 0.14

fusion may be concealed by the use of a Gaussian model. As

G4

such, the MD may not be a good choice for characterizing

the changes in tumor histopathology or, consequently, for
tumor grading. The MD, MK, and Krad were useful in the
discrimination of G1 vs G3, G1 vs G4, and G2 vs G4, in-
dicating that although the kurtosis metrics can provide added
375.17 ± 120.88

MK, mean kurtosis; NA, not applicable; N/C, nuclear-to-cytoplasm ratio.

value or new information beyond what can be inferred from

1.92 ± 0.70
1.53 ± 0.39
1.50 ± 0.32
1.33 ± 0.42
1.45 ± 0.64
2.75 ± 0.88
0.32 ± 0.09
0.54 ± 0.13

0.61 ± 0.14

the conventional diffusivity metrics, the differences may not

G3

reach statistical significance in circumstances in which there

is a limited number of subgroup cases. In fact, variations in
molecular crowding and viscosity among different intercel-
lular components would affect the MD at the intracellular level
(23).
397.54 ± 85.710

The Kax is defined as the kurtosis measured along the di-

2.75 ± 0.49
0.97 ± 0.30
0.94 ± 0.31
1.13 ± 0.31
2.39 ± 0.45
3.49 ± 0.61
0.29 ± 0.17
0.33 ± 0.11

0.63 ± 0.18

rection parallel to the principal diffusion direction (ie, the first

G2

diffusion eigenvector), whereas the Krad is defined as the

average kurtosis measured in the plane perpendicular to the
first eigenvector. The MK is the average of the kurtosis along
all directions of a densely sampled sphere (24). The MK, Kax,
377.67 ± 191.38

and Krad can reflect the degree of diffusional kurtosis, yet the
2.80 ± 0.55
0.90 ± 0.31
0.81 ± 0.29
1.03 ± 0.27
2.41 ± .049
3.60 ± 0.71
0.30 ± 0.07
0.24 ± 0.09

0.65 ± 0.22

physical meaning of the MK, Kax, and Krad relative to cell

G1

properties and histologic details remains a matter of debate,

as the precise value of each parameter should be different. In
our study, compared to the MD, the MK and Krad were better
at differentiating G2 vs G3, whereas the MK and Kax were
better at differentiating G3 vs G4, which may be attributed
Drad
Krad

CNC
CVF
Dax

N/C
Kax
MD

to the fact that the MK can reflect the specific aspect of mo-
MK

FA

lecular water diffusion known as “complexity,” which could

434
Downloaded for Siti Fadhilah (fadhilahhazhiyah@gmail.com) at Universitas Muslim Indonesia from ClinicalKey.com by Elsevier on August 30, 2018.
For personal use only. No other uses without permission. Copyright ©2018. Elsevier Inc. All rights reserved.
Academic Radiology, Vol 25, No 4, April 2018 DKT IMAGING AND DWI IN CCRCC

TABLE 3. Comparative Analysis of Diffusion and Kurtosis Metrics Between Tumor and Normal Parenchyma and Between
Tumors with Different Grades

MD MK Krad Kax Drad Dax FA

Cortex vs medulla 0.880 0.766 0.710 0.696 0.887 0.824 0.712
AUC
P Value <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 <0.001
ccRCC vs cortex 0.910 0.842 0.818 0.924 0.893 0.899 0.538
AUC
P Value <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 0.379
ccRCC vs medulla 0.587 0.661 0.616 0.819 0.521 0.707 0.681
AUC
P Value 0.043 <0.001 0.007 <0.001 0.617 <0.001 <0.001
ccRCC (G1 vs G2) 0.535 0.585 0.634 0.640 0.519 0.503 0.636
AUC
P Value 0.640 0.258 0.064 0.077 0.966 0.799 0.072
ccRCC (G1 vs G3) 0.846 0.895 0.946 0.686 0.870 0.806 0.581
AUC
P Value <0.001 <0.001 <0.001 0.044 0.001 0.001 0.383
ccRCC (G1 vs G4) 0.954 0.981 0.981 0.968 0.958 0.889 0.579
AUC
P Value <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 0.492
ccRCC (G2 vs G3) 0.799 0.898 0.904 0.608 0.821 0.736 0.662
AUC
P Value <0.001 <0.001 <0.001 0.203 <0.001 0.005 0.055
ccRCC (G2 vs G4) 0.919 0.967 0.950 0.944 0.953 0.865 0.617
AUC
P Value <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 0.279
ccRCC (G3 vs G4) 0.639 0.796 0.650 0.839 0.683 0.589 0.528
AUC
P Value 0.231 0.022 0.195 0.010 0.114 0.443 0.811

AUC, area under curve; ccRCC, clear cell renal cell carcinoma; Dax, axial diffusivity; Drad, radial diffusivity; FA, fractional anisotropy; Kax,
axial kurtosis; Krad, radial kurtosis; MD, mean diffusivity; MK, mean kurtosis.

TABLE 4. The Correlation of the Diffusion and Kurtosis Metrics with Tumor Histopathology

MD MK Krad Kax Drad Dax FA

N/C
r −0.474 0.543 0.560 0.438 −0.507 −0.433 −0.092
P <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 0.387
CNC
r −0.074 0.022 0.026 −0.019 −0.075 −0.073 −0.081
P 0.487 0.834 0.804 0.862 0.483 0.490 0.447
CVF
r 0.039 −0.050 0.005 −0.053 0.045 0.014 −0.066
P 0.715 0.637 0.961 0.615 0.669 0.892 0.534

CNC, cell nuclei count; CVF, cell volume fraction; Dax, axial diffusivity; Drad, radial diffusivity; FA, fractional anisotropy; Kax, axial kurto-
sis; Krad, radial kurtosis; MD, mean diffusivity; MK, mean kurtosis; N/C, nuclear-to-cytoplasm ratio.

be caused by bizarre, multilobed nuclei with heavy chroma-
tin clumps and increased areas of spindle cells in G4 that did
not exist in G3 (25), as well as increasing complications of
the cytoplasmic constituents with increasing grades. Al-
though the performance of kurtosis in different directions may
differ, the MK in the evaluation remains stable. It is well rec-
ognized that the diffusion directions inside tumors are highly
heterogeneous (26). We posit that the MK should exhibit the
best performance in tumor diagnosis, whereas the Kax and
Krad are defined to specifically address the direction-
dependent kurtosis and would lose generality for tumor cases.
The FA lacked significance in differentiating tumor grades.
The known fact that the FA in the kidneys depends on the
patient age and gender (27) may act as a confounder in the
evaluation of the FA for grading ccRCC. Meanwhile, the re-
striction of anisotropy in diffusion to similar degrees in the

435
Downloaded for Siti Fadhilah (fadhilahhazhiyah@gmail.com) at Universitas Muslim Indonesia from ClinicalKey.com by Elsevier on August 30, 2018.
For personal use only. No other uses without permission. Copyright ©2018. Elsevier Inc. All rights reserved.
WU ET AL Academic Radiology, Vol 25, No 4, April 2018

Figure 3. Receiver operating characteristic plots of diffusivity and kurtosis metrics in different clear cell renal cell carcinoma grades. (a)
G1 vs G2, (b) G1 vs G3, (c) G1 vs G4, (d) G2 vs G3, (e) G2 vs G4 and (f) G3 vs G4. Dax, axial diffusivity; Drad, radial diffusivity; Kax, axial
kurtosis; Krad, radial kurtosis; MD, mean diffusivity; MK, mean kurtosis.

436
Downloaded for Siti Fadhilah (fadhilahhazhiyah@gmail.com) at Universitas Muslim Indonesia from ClinicalKey.com by Elsevier on August 30, 2018.
For personal use only. No other uses without permission. Copyright ©2018. Elsevier Inc. All rights reserved.
Academic Radiology, Vol 25, No 4, April 2018
solid region of different grades could also be an appropriate
explanation for the result.
However, the MK was shown to be a promising imaging
marker for this differentiation.
The differences in the diffusivity and kurtosis metrics between
ccRCC and normal renal parenchyma were also assessed. The
MK, Krad, and Kax values were higher in the ccRCC tissues
than in the normal medulla and cortex, and it is easily un-
derstandable that a malignant tumor has a more complex
structure that could be reflected by the kurtosis metrics. For
the diffusivity metrics, in agreement with a previous study (15),
significantly lower FA, MD, and Dax values, as well as lower
Drad values, were found in the ccRCC than in the normal
renal cortex. In contrast, we found no obvious difference in
the Drad between the ccRCC and the normal medulla; ac-
cording to previous studies, it is possible that radially oriented
vessels, tubules, and collecting ducts in the medulla contrib-
ute to the decreased diffusivity metrics, potentially explaining
our results (28–30). In addition, regarding the differences
between the normal cortex and the medulla, the kurtosis metrics
were higher in the medulla than in the cortex, which was
compatible with the knowledge that the renal medulla has a
more complex structure and that the restriction of water dif-
fusion by the walls of collecting tubules and ducts greatly
deviates the diffusion distribution from the Gaussian form. We
also found higher FA and lower MD, Drad, and Dax values
in the medulla than in the cortex, which may contribute to
corticomedullary differences in the structures of the vessels,
tubules, and collecting ducts.
Because histologic examination is the standard reference for
ccRCC diagnosis and grading, the diagnosis of ccRCC using
diffusivity and kurtosis metrics was also evaluated in terms of
tumor histopathology. A previous study (11) that explored
the ability of the MD and MK to detect diffusion properties
among phantoms with different degrees of structural changes,
demonstrated that the MK may be a more specific indicator
for tissue structural complexity, which is not reflected by the
MD. Thus, we further explored the relationship between kur-
tosis metrics and ccRCC tumor histopathology features.
According to the Fuhrman grading system (25), nuclei in higher
tumor grades are larger and show a more irregular appear-
ance, leading to increases in the N/C ratio with increasing
tumor grade. However, the CVF did not change dramati-
cally, suggesting that the molecular motion of water becomes
more restricted because of decreasing extranuclear space as
tumor cells proliferate and microstructural complexity in-
creases within the tumor. The correlation between the kurtosis
metrics and the N/C ratio may reflect physiological and mor-
phologic alterations associated with ccRCC because kurtosis
metrics may be related to the degree of microstructural com-
plexity, indicating microstructural changes in ccRCC. However,
it would be premature to conclude a cause-and-effect rela-
tionship. The mechanisms underlying the differences in the
alterations of kurtosis metrics with ccRCC grades have yet
to be determined, and the relationship between restricted water
diffusion and microstructural complexity remains unclear.
Downloaded for Siti Fadhilah (fadhilahhazhiyah@gmail.com) at Universitas Muslim Indonesia from ClinicalKey.com by Elsevier on August 30, 2018.
For personal use only. No other uses without permission. Copyright ©2018. Elsevier Inc. All rights reserved.
DKT IMAGING AND DWI IN CCRCC
In our study, 600 s/mm2 was chosen as the highest b value
in our DWI protocol. Higher b values probed in DKT imaging
reflect a different population of restricted spins (diffusing within
smaller pores), and the diffusion properties of non-Gaussian
factors become more significant as the b value increases. As
the K value represents a deviation from a Gaussian model of
water motion, using DTI data collected with high b values
(within the appropriate range of maximum b values for DKT
imaging) should increase the robustness of the K estimates.
Despite the fact that in DWI (low b values), the diffusion is
assumed to be approximated by the first terms of the Taylor
or cumulant expansions of the signal, and as such, diffusion
in low b value can be considered “Gaussian,” and the fact is
implicit that they will reflect restriction as will the kurtosis
tensor, although it will be the second-order term in the ex-
pansions. Moreover, using lower b values can save time and
provide a better signal-to-noise ratio for DWI, especially when
using a 1.5-T scanner. However, whether DKT imaging with
higher b values (probing smaller dimensions) will provide ad-
ditional information is unknown and requires further
exploration.
There are some limitations to our study. First, the study
was performed using a 1.5-T scanner. Although a recent study
demonstrated that the diffusivity and kurtosis metrics mea-
sured with DKI at 1.5 and 3.0 T are strongly correlated and
typically differ by only a few percent, which supports the ro-
bustness of the DKI results with respect to field strength in
the human brain (31), its validation for application in ccRCC
requires further investigation. Second, we did not compare
our results with the prior fast diffusion kurtosis approach (18).
However, as the literature (32) has demonstrated good agree-
ment between the MK derived from a tensor method and a
nontensor method in an animal model of acute stroke, we
believe it can be used for reference to the comparison between
the tensor method used in our study and the nontensor method
used in the former study (18). Moreover, nuclear grading was
based on the highest-grade area identified within the tumor
but was dependent on the extent of sampling of a given tumor.
Further research focusing on this topic is needed to explore
the true value of DKT imaging in discriminating different grades
of ccRCC.
CONCLUSION
DKT imaging is feasible for demonstrating the non-
Gaussian behavior of water diffusion in ccRCC and provides
better performance than conventional DWI in grading ccRCC,
which demonstrated that DKT could be a promising method
for characterizing ccRCC in terms of the tumor microenvi-
ronment and reflecting both the microstructural changes
and complexity of the tumor tissue. We hope that our study
will serve as an extension of previous studies on normal
kidneys and provide preliminary information for the further
study of applying this new technique in various malignant
and benign renal masses, as well as in differentiating RCC
subtypes.
437
WU ET AL
REFERENCES
1. Cheville JC, Lohse CM, Zincke H, et al. Comparisons of outcome and
prognostic features among histologic subtypes of renal cell carcinoma.
Am J Surg Pathol 2003; 27:612–624. doi:10.1097/00000478-200305000-
00005.
2. Gill IS, Remer EM, Hasan WA, et al. Renal cryoablation: outcome at 3
years. J Urol 2005; 173:1903–1907. doi:10.1097/01.ju.0000158154
.28845.c9.
3. Jocham D, Richter A, Hoffmann L, et al. Adjuvant autologous renal tumour
cell vaccine and risk of tumour progression in patients with renal-cell car-
cinoma after radical nephrectomy: phase III, randomised controlled trial.
Lancet 2004; 363:594–599. doi:10.1016/S0140-6736(04)15590-6.
4. Wesbey GE, Moseley ME, Ehman RL. Translational molecular self-
diffusion in magnetic resonance imaging. II. Measurement of the self-
diffusion coefficient. Invest Radiol 1984; 19:491–498. doi:10.1097/
00004424-198411000-00005.
5. Le Bihan D, Breton E, Lallemand D, et al. MR imaging of intravoxel in-
coherent motions: application to diffusion and perfusion in neurologic
disorders. Radiology 1986; 161:401–407. doi:10.1148/radiology
.161.2.3763909.
6. Basser PJ. Inferring microstructural features and the physiological state
of tissues from diffusion-weighted images. NMR Biomed 1995; 8:333–
344. doi:10.1002/nbm.1940080707.
7. Bonekamp S, Corona-Villalobos CP, Kamel IR. Oncologic applications
of diffusion-weighted MRI in the body. J Magn Reson Imaging 2012;
35:257–279. doi:10.1002/jmri.22786.
8. Zhang J, Tehrani YM, Wang L, et al. Renal masses: characterization with
diffusion-weighted MR imaging—a preliminary experience. Radiology 2008;
247:458–464. doi:10.1148/radiol.2472070823.
9. Wang H, Cheng L, Zhang X, et al. Renal cell carcinoma: diffusion-
weighted MR imaging for subtype differentiation at 3.0 T. Radiology 2010;
257:135–143. doi:10.1148/radiol.10092396.
10. Yu X, Lin M, Ouyang H, et al. Application of ADC measurement in char-
acterization of renal cell carcinomas with different pathological types and
grades by 3.0 T diffusion-weighted MRI. Eur J Radiol 2012; 81:301–
306.
11. Van Cauter S, Veraart J, Sijbers J, et al. Gliomas: diffusion kurtosis MR
imaging in grading. Radiology 2012; 263:492–501. doi:10.1148/
radiol.12110927.
12. DeCarlo LT. On the meaning and use of kurtosis. Psychol Methods 1997;
2:292–307. doi:10.1037/1082-989X.2.3.292.
13. Jensen JH, Helpern JA. MRI quantification of non-Gaussian water dif-
fusion by kurtosis analysis. NMR Biomed 2010; 23:698–710.
doi:10.1002/nbm.1518.
14. Rosenkrantz AB, Sigmund EE, Winnick A, et al. Assessment of hepa-
tocellular carcinoma using apparent diffusion coefficient and diffusion
kurtosis indices: preliminary experience in fresh liver explants. Magn Reson
Imaging 2012; 30:1534–1540. doi:10.1016/j.mri.2012.04.020.
15. Pentang G, Lanzman RS, Heusch P, et al. Diffusion kurtosis imaging of
the human kidney: a feasibility study. Magn Reson Imaging 2014; 32:413–
420. doi:10.1016/j.mri.2014.01.006.
16. Huang Y, Chen X, Zhang Z, et al. MRI quantification of non-Gaussian
water diffusion in normal human kidney: a diffusional kurtosis imaging
study. NMR Biomed 2015; 28:154–161. doi:10.1002/nbm.3235.
438
Downloaded for Siti Fadhilah (fadhilahhazhiyah@gmail.com) at Universitas Muslim Indonesia from ClinicalKey.com by Elsevier on August 30, 2018.
For personal use only. No other uses without permission. Copyright ©2018. Elsevier Inc. All rights reserved.
Academic Radiology, Vol 25, No 4, April 2018
17. Tamura C, Shinmoto H, Soga S, et al. Diffusion kurtosis imaging study
of prostate cancer: preliminary findings. J Magn Reson Imaging 2014;
40:723–729. doi:10.1002/jmri.24379.
18. Dai Y, Yao Q, Wu G, et al. Characterization of clear cell renal cell car-
cinoma with diffusion kurtosis imaging: correlation between diffusion
kurtosis parameters and tumor cellularity. NMR Biomed 2016; 29:873–
881. doi:10.1002/nbm.3535.
19. Jiang R, Jiang J, Zhao L, et al. Diffusion kurtosis imaging can efficiently
assess the glioma grade and cellular proliferation. Oncotarget 2015;
6:42380–42393. doi:10.18632/oncotarget.5675.
20. Shrout PE, Fleiss JL. Intraclass correlations: uses in assessing rater
reliability. Psychol Bull 1979; 86:420–428. doi:10.1037/0033-2909.86.2
.420.
21. Büsing KA, Kilian AK, Schaible T, et al. Reliability and validity of MR image
lung volume measurement in fetuses with congenital diaphragmatic hernia
and in vitro lung models. Radiology 2008; 246:553–561.
doi:10.1148/radiol.2462062166.
22. Rosenkrantz AB, Niver BE, Fitzgerald EF, et al. Utility of the apparent
diffusion coefficient for distinguishing clear cell renal cell carcinoma of
low and high nuclear grade. AJR Am J Roentgenol 2010; 195:W344–
W351. doi:10.2214/AJR.10.4688.
23. Le Bihan D. Apparent diffusion coefficient and beyond: what diffusion
MR imaging can tell us about tissue structure. Radiology 2013; 268:318–
322. doi:10.1148/radiol.13130420.
24. Wu EX, Cheung MM. MR diffusion kurtosis imaging for neural tissue
characterization. NMR Biomed 2010; 23:836–848. doi:10.1002/
nbm.1506.
25. Fuhrman SA, Lasky LC, Limas C. Prognostic significance of morpho-
logic parameters in renal cell carcinoma. Am J Surg Pathol 1982; 6:655–
663. doi:10.1097/00000478-198210000-00007.
26. Diaz-Cano SJ. Tumor heterogeneity: mechanisms and bases for a reli-
able application of molecular marker design. Int J Mol Sci 2012; 13:1951–
2011. doi:10.3390/ijms13021951.
27. Chan RW, Von Deuster C, Stoeck CT, et al. High-resolution diffusion tensor
imaging of the human kidneys using a free-breathing, multi-slice, tar-
geted field of view approach. NMR Biomed 2014; 27:1300–1312.
doi:10.1002/nbm.3190.
28. Thoeny HC, de Keyzer F, Oyen RH, et al. Diffusion-weighted MR imaging
of kidneys in healthy volunteers and patients with parenchymal dis-
eases: initial experience. Radiology 2005; 235:911–917. doi:10.1148/
radiol.2353040554.
29. Ries M, Jones RA, Basseau F, et al. Diffusion tensor MRI of the
human kidney. J Magn Reson Imaging 2001; 14:42–49. doi:10.1002/
jmri.1149.
30. Gürses B, Kiliçkesmez O, Taşdelen N, et al. Diffusion tensor imaging of
the kidney at 3 Tesla MRI: normative values and repeatability of mea-
surements in healthy volunteers. Diagn Interv Radiol 2011; 17:317–322.
doi:10.4261/1305-3825.DIR.3892-10.1.
31. Shaw CB, Jensen JH, Deardorff RL, et al. Comparison of diffusion
metrics obtained at 1.5T and 3T in human brain with diffusional kurto-
sis imaging. J Magn Reson Imaging 2017; 45:673–680. doi:10.1002/
jmri.25380.
32. Sun PZ, Wang Y, Mandeville E, et al. Validation of fast diffusion kurto-
sis MRI for imaging acute ischemia in a rodent model of stroke. NMR
Biomed 2014; 27:1413–1418. doi:10.1002/nbm.3188.