IJISET - International Journal of Innovative Science, Engineering & Technology, Vol. 2 Issue 4, April 2015.
www.ijiset.com
ISSN 2348 – 7968
4D Image Analysis and Diagnosis of Kidney Disease Using DCE-MRI
Images
Nikita Derle, Prof. Devidas Dighe
Department of Electronics and Telecommunication, Savitribai Phule Pune University,
Matoshri College of Engineering and Reseach Center, Nasik, India
Abstract
Because of noninvasive nature, medical imaging is easy to
perform though it is extravagant. For furnishing superior
anatomy and decisiveness, different characteristics have been
extrapolated from intake image. Earlier the processing steps like
registration, segmentation are separately applied for extraction of
sequential proprieties of DCE-MRI images of kidney. For
simultaneous registration and segmentation of the kidney, a 4D
model is described. In the conscript of kidney abnormal
functioning and disease detection, the glomerular filtration rate
(GFR) is a significant factor. Dynamic contrast enhancement
magnetic resonance imaging (DCE-MRI) is the imaging
proficiency, used for calibrating different parameters
homologous to suffuse, capillary leakage, and convey rate in
tissues of various organs and diseases detection. The described
technique’s approach permits us to automatically accomplishing
a statistical analysis of various parameters from alive cells.
Conclusion of findings is accomplished by average gray level
intensity inside the kidney region.
Keywords: DCE-MRI, Kidney, Registration, Segmentation,
Renal Function, GFR, Spatial mapping, CKD.
1. Introduction
Dynamic
Contrast
Enhanced
Magnetic
Resonance Imaging (DCE-MRI) is used by
radiologist for diagnosis of antithetic
contaminations.
For
a
little
while,
pronouncement is concluded through biopsy,
which may expedite bleeding and infections in
the humanitarian. Teensy irk biopsies may
effectuate overestimation or underestimation of
the excess of inflammation. In DCE-MRI, GDDTPA is a contrast chemical agent injected into
the patient’s bloodstream to perfuse inside ilk,
precipitated in recreation of flush, to catalyze a
discrepancy image of organ. The working facts
and MRI gives analyzed data, habituated to
catch out affections.
DCE-MRI of the kidney has advantages: i)
Radiation exposure is not required ii) Three
dimensional acquisitions are used iii) spatial
resolution is superior.[18] Accompanied by
difficulties like: i) low spatial resolution due to
fast scanning, ii) selection of proper registration
process, iii) choosing correct segmentation
process. In the ilk debilitation, GFR is the
foremost parameter. GFR is measured from the
blood pool in the glomerular capillaries of the
volume of filtered fluid per unit time. If the
value of GFR is low, then there are chances of
diseases and dysfunction of the breed. [20]
In the existing system, there is no focus on the
voxel based GFR measurement from DCE-MRI
acquisitions. In 3D model optimization problem
is observed. Registration and segmentation move
are not associate together. This model did not
contribute for diagnosis. In the proposed system,
these problems can be overcome by using a 4D
model rather than 3D model. 4D model fragment
the ascription by catering registration and
segmentation. In 4D model 4 steps convoluted
are
(i)
Registration.
(ii)
Segmentation.
(iii) Compartment modeling.
(iv)
Combined
of
registration
and
segmentation.
1.1 Risk factors diagnosis and classification:
Kidney affliction appertains to any renal
pathology that has implied to antecedent
deduction in renal operative interact. This is
most imposing associate with a reduction in
GFR but other imperial functions may be mislaid
within this to be found. [22]
Diabetic Mellitus: Diabetic nephropathy is a
renal complexity of diabetic mellitus. The
association of diabetic with inopportune bear of
290
IJISET - International Journal of Innovative Science, Engineering & Technology, Vol. 2 Issue 4, April 2015.
www.ijiset.com
ISSN 2348 – 7968
chronic kidney disease (CKD) is more
excruciating to substantiate. In one cross
sectional study diabetes was relative imminence
aggrandizing with the despotic of CKD.
Hypertension: It is a hazard aspect of CKD.
1.2
Detecting Kidney Damage:
Kidney damage may be detected directly or
indirectly by using imaging or hostopathelogical
examination of renal biopsy direct evidences
found out. Imaging types used are computed
tomography (CT), magnetic resonance imaging
(MRI), ultrasound, and isotope scanning may
detect number of structural monstrosity. The
structural abnormalities include polycystic
kidney disease, reflux nephropathy, chronic
pyelenephritis and renovascular disease. From
urinalysis, indirect evidence may be inferred for
kidney damage. [23].
2. Literature Review
Literature survey is able to critically summarize the
current technology in the area of nephrology for any
potency and debilitation in precedent work. Identify
the technique to eliminate the implicit wimpiness,
whilst bringing to the fore the potential strengths.
a) Reviews on kidney tests and diagnosis
techniques: Blood and urine tests can be performed
to check for the kidney function. In order to diagnose
and confirm disorders related to kidney blood vessels,
tolerant can endure kidney biopsy proceeding.
Imaging tests are performed to get useful information
about kidney structures. [22] Blood tests: commonly
used tests are blood urea nitrogen and serum
creatinine from the blood samples. The concentration
determination of these two substances is not sensitive
enough as the concentration will not exceed the
normal reference ranges until there is loss of more
than 75% of kidney function. Urine tests: The
degree of kidney impairment can be assessed by
measuring the GFR, and information on the cause of
the disease can be achieved by urine test. Imaging
tests: to get useful information regarding the kidney
structures using different imaging modalities.
Fredrik Maes et al [1] propose images registration
using maximization of mutual information (MMI) for
CT and MRI images. The steps used are non-rigid
image matching, rectification, shape normalization,
motion estimation, tissue deformation correction all
are the area to be explored. Macros MartinFernandez et al [2] developed an approach for
contour detection of human kidneys from ultrasound
images using Markov random fields and active
contours. It is a probabilistic Bayesian method.
Segmentation of a vivo kidney out of volumetric
series of 2D echo graphical slices. But quality of
solution is a problem with ultrasound imaging. Ali
Gholipour et al [3] provided classification of the
image registration techniques for CT and MRI
images for proper selection based on resolution.
Asem M. Ali et al [4] explore the idea on graph cuts
frameworks by using segmentation with proper shape
constraints of the kidney. Provide results better than
clinical results but with manual inputs. Giele et al [6]
ameliorated
the
antecedent
techniques
by
appertaining erosion to the mask image to acquire a
contour via a second deduction stage. Several rings
were obtained, which formed the basics of the
segmentation of the cortex from the medulla
structures. Boykov et al [7] used graph cuts to get a
across-the-board optimal intention extraction
approach for dynamic N-data sets for minimized cost
function. Although the results appeared auspicious,
manual interaction was still challenge. Priester et al
[5] abated the average of pre-contrast images by
using generated threshold from the average of earlyenhancement images, and black-and-white kidney
mask. This mask image is corroded and the kidney
contour is acquired with help of manual interactions.
Sun et al [8] introduced numerous computerized
artifices for kidney segmentation and registration.
Aly A. Farag et al [9] offered an efficacious come up
for the shape based segmentation problem using level
sets. It is based on dissimilarity scales by using a
dissimilarity measure approximated by a smeared
version of the maximum function using DCE-MRI
images. Frank G. Zollner [10] et al explored idea on
image analysis methods in the assessment of human
kidney perfusion based on 3D DCE-MRI data. Found
the K-means clustering is a suitable approach for
time course analysis of renal perfusion when proper
motion correction is performed as a preprocessing
step, but it deals with the processing of observed
291
IJISET - International Journal of Innovative Science, Engineering & Technology, Vol. 2 Issue 4, April 2015.
www.ijiset.com
ISSN 2348 – 7968
signal time intensity courses only. Tilo W. Eichler et
al [11] explore the automated cell segmentation
represents a robust and powerful tool for the
statistical analysis of various cell parameters. Toufik
Sari [12] provided several binarization techniques
with feature extraction. Nick method is the best
binarizaton method among all methods. Volker
Daum et al [13] proposed work on segmentation of
kidneys using a new active shape model generation
techniques based on non-rigid image registration
with curvature-based image registration gives an
attractive alternative to minimum description length
(MDL) based techniques. Davies et al [14] proposed
a description of an automatic method for the
construction of optimal 3D statistical shape models.
V Rajamani et al [15] worked on comparison of local
binary pattern variants for ultrasound kidney image
retrieval. The efficiency of the system is measured
using recall and precision parameter. S. Manikandan
[16] analyzed various features of an ultrasound
kidney images using gray level co-occurrence
matrices. Jeff L. Zhang et al [17] characterizes the
renal functions presents in MRI of the kidney.
Limitations of the tools are robust image registration
and segmentation, broad area of tools for MRI
interpretation of Blood Oxygen Level Dependent
(BOLD) measurement, also the challenges with
respiratory and bulk motion of tissues are critical to
measure in function renal MRI. Junchi Tokuda et al
[18] worked on non-rigid motion segmentation with
motion correction techniques on pixel wise
pharmacokinetic analysis of free breathing
pulmonary DCE-MRI. It is for therapy response only.
Louisa Bokacheva et al [19] estimate on GFR from
MR Renography and tracer kinetic models. 4D or 5D
model can provide RPF but with high SNR is the
limitation of this system. A. M. Khan et al [20]
provided various segmentation methods. It is hard to
obtain single answer for segmentation of given
images as the interpretation varies from individual
approaches. Hodneland et al [21] introduced the idea
of combination of registration and segmentation
which is applicable to the 4D DCE-MRI of moving
human kidney. The limitations to this system are
GFR measurement and diagnosis of renal diseases.
Yetzi et al enhanced his work on segmentation and
registration in active contours of 2D images. For nonaffine registration this method is not suited. From
survey it is found that various imaging modalities,
segmentation and registration techniques are
proposed. Features selection is a critical issue and
need to resolve. There is need to combine the
segmentation and registration process and identify
the proper features for automatic kidney disease
diagnosis using suitable imaging modality. With this
intention we have proposed a new MRI renography
approach for diagnosis of various renal diseases and
GFR value measurement and optimization of image.
Proposed system is an integral model with the
combination of registration and segmentation.
3. Proposed System
Fig. 1 General Block Diagram of the Proposed System
In the existing system there is a huge work
carried out on 2D and 3D model which contains
registration,
segmentation
and
compartment
modeling but it consists of optimization problem.
Manual segmentation included most of the system so
there is problem of handshaking errors. Image
registration and segmentation provided are not robust
to the noise. Also GFR is not computed previously
which is an important parameter for filtration rate
calculation. Acute rejection and normal patients
classification is not possible. There is no approach
toward diagnosis of renal diseases in the past.
In the proposed system, the optimization
problem can be overcome. Handshaking errors can be
removed by automatic segmentation. It is the
combination of simultaneous registration and
segmentation, also calculates the GFR value by using
voxel deformations. Also boundary area is separated
by using the spatial mapping and segmentation on the
area of the image.
Set of some standard rules can be defined for the
diagnosis of various renal diseases of the kidney. In
4D model, proposed system extracts the features by
combining registration and segmentation.
292
IJISET - International Journal of Innovative Science, Engineering & Technology, Vol. 2 Issue 4, April 2015.
www.ijiset.com
ISSN 2348 – 7968
i) Registration: Procedures of registration
classification depend on the registration paradigm.
Features are either extrinsic or intrinsic. Extrinsic
method is related to external objects or markers
introduced in the image space. Intrinsic method can
be point based or surface based or voxel based.
Voxel based registration method optimize a
functional measuring the similarity of all
geometrically corresponding voxel pairs for some
features.[1] Anatomical objects extracted from 3D
medical images are aligned using affine
transformations to remove the global size differences.
The affine transform T of point p = (p1,…..,pd)’ Є Rd
to q = (q1,……,qd)’ is given by
q = Rp + c
(1)
Where the matrix R corresponds to rotation,
scaling and shear and c corresponds to translation.
ii) Segmentation: approach behind the
segmentation is to simplify image which is more
meaningful and easier for analysis purpose. The
system gets contours of extracted image or a set of
segments which covers the whole area of an image
from
this
segmentation
procedure.
Image
segmentation is the fundamental step to analyze
images and extract data from them. Region growing
method has better performance for DCE-MRI
images. [9]
Algorithm for region growing method
Let R represent the entire image region and let p be
any predicate.
If P(R) = False
Divide image into quadrants
If p is possible for any sub quadrant
Subdivide that quadrant into sub quadrant.
Stop dividing when P is true.
Merge the regions Rj & Rk (j ≠ k; j = 1,2,3,…..,n, k =
1,2,3,….,n)
if P (Rj Є Rk) = = true
iii) Compartment Modeling: There are one or
two compartment models formed in the system which
has parts of tissues, as the contrast agent gets
exchanged. For DCE-MRI images, these models are
more simplistic. If models have more than two
compartments, redundancy issue gets developed.
Because of these redundancy issues system cannot
find model parameters which are important. If the
model is more complex there is regression problem
arises. However, a spatial elastic approach can be
used which provides number of compartments for
each voxel so that model complexity is not fixed. As
we use a spatial elastic approach, the system gets a
sparse set of basic functions for each voxel, so that
the rate gets constant in each compartment. For
simulated images this method is used and it can be
applied to vivo datasets also. [21]
iv) Combined Registration and Segmentation:
For combination of registration and segmentation,
use elasticity regularizes and distance measure. In
between template and reference image there is a
smooth deformation. 3D images are formed by using
sequentially parallel 2D images, in which a smallest
element present has a cubic volume is called a voxel.
3D images have limited spatial resolution, minor
artifacts and gray scale, to avoid its effect before
image viewing, image filtering should be done.
Patients’ anatomy and physiology are given by 3D
images. 4D images are formed by using temporal
series of 3D images. 4D images represent patient’s
motion over time. Patient’s motion can be faster or
slower. Faster motion related with speed which
causes blurring artifacts and slower motion may not
be affected on the image quality.
Figure 1 shows the general approach of the proposed
system. In the system DCT-MRI image is an input
which is first preprocessed. Then carry out
optimization to overcome Optimization problem.
Voxel deformation is done after the optimization of
the image for GFR calculation. Spatial mapping is
required for boundary detection to differentiate
between kidney and other than kidney area for
accurate segmentation of kidney wall regions.
System will compute the GFR from Voxel
deformation’s RMSE measurement for kidney
region. Segmentation is required for kidney wall and
remaining area of the image for both kidneys
separately to avid errors. Boundary of kidney wall is
detected by using canny edge detector. After that
carry out analysis by using some set of standard rules
suggested by radiologist, which is required for
diagnosis of the renal diseases.
293
IJISET - International Journal of Innovative Science, Engineering & Technology, Vol. 2 Issue 4, April 2015.
www.ijiset.com
ISSN 2348 – 7968
4. Conclusions
Figure 2: (a) Original Image after Preprocessing (b) After Gaussian
Filtering (c) Result of Thresholding (d) Result of Boundary Detection
After Thresholding.
The proposed system will overcome the problems
with 3D model due to moving parts. GFR
measurement is done by using voxel deformation
which is an important parameter in diagnosis. More
accurate boundary detection is possible as carried out
after segmentation. Finally diagnosis of renal
diseases is expected by using the standard sets of
rules are still to be defined. Results of some
processing steps are shown for the system in figure 2.
Acknowledgments
References
[1].
F.
Maes,
A.
Collignon,
D.
Vandermeulen,
G.
Marchal,
and
P.
Suetens,
“Multimodality
image
registration by maximization of mutual
information,” IEEE Trans. Med. Imag.,
vol. 16, no. 2, pp. 187–198, 1997.
[2].
M. M. Fernandez, C. A. Lopez, “An
approach for contour detection of human
kidneys from ultrasound images using
Markov
random
fields
and
active
contours,” Elsevier B.V. 1361/8415/$,
2004.
[3].
A. Gholipur, N. Kehtarnavaz, “Brain
functional localization: A survey of
image
registration
techniques,”
IEEE
Trans. Med. Imag., vol. 26, no. 4, 2007.
[4].
A. Ali, A. A. Farag, and A. S. ElBaz, “Graph cuts framework for kidney
segmentation
with
prior
shape
constraints,” LNCS 4791, pp. 384-392,
Springer-Verlag Berlin Heidelberg, 2007.
[5].
De Priester, J. Kessels, A. Giele, E.
den Boer, J. Christiaans, M. Hasman, A.
van Engelshoven, J. ” MR Renography by
semi
automated
image
analysis:
performance
in
renal
transplant
recipients.
J.
Magn.
Reson
Imaging
14(2), 134-140 2001.
[6].
Giele
E.
“Computer
methods
for
semiautomatic
MR
Renogram
determination,” PhD thesis, Eindhoven
University
of
Technology,
Eindhoven
2002.
[7].
Boykov, Y. Lee, V. S. Rusinek, H.
Bansal, R. “Segmentation of dynamic N-D
data sets via graphs cuts using markov
models. In. Niessen, W. J. Viergever, M.
A. MICCAI. LNCS, vol 2208, pp. 10581066,” Springer, Heidelberg, 2001.
[8].
Sun, Y., Jolly, M. P., Mourua, J. M.
“Integrated
registration
of
dynamic
renal perfusion MR images,” In. Proce.
IEEE ICIP, pp. 1923-1926. IEEE Computer
Society Press, Los Alamitos, 2004.
[9].
A.
A.
Farag,
W.
Miller,
H.
Abdelmunim, M. aboelGhar, “A kidney
segmentation approach from DCE-MRI using
level sets,” IEEE 978-1-4244-2340-8/08,
2008.
[10]. F. G. Zollner, R. Sance, P. Rogelj,
M. J. Lesma-Carbayo, J. Rorvik, A.
Santos, A. Lundervold, “Assessment of 3D
DCE-MRI of the kidneys using non-rigid
image registration and segmentation of
voxel time courses,” Elsevier, 2008.
[11]. N. V. Bukoreshtliev, T. W. Eichler,
E. Hodneland, Xen-Cheng Tai, S. Gurke,
A. Lundervold, “A unified framework for
automated 3D segmentation of surfacestained living cells and a comprehensive
segmentation evaluation,” IEEE Trans.
Med. Imag., vol 28, no. no. 5, 2009.
[12]. A. Kefali, T. Sari, M. Sellami,
“Evaluation
of
several
binarization
techniques for old arabik documents
images,” University BadjiMokhtar, 2009.
[13]. M. Spigel, D. A. Hahn, V. Daum, J.
Wasza, J. Hornegger, “Segmentation of
kidneys using a new active shape model
generation based on non-rigid image
registration,” Elsevier, 2008.
[14]. D. R. Twining C, Cootes T, Waterton
J, Taylor C, “3D statistical shape
models using direct optimization of
description
length,”
In:
European
conference on computer vision, vol. 3,
p.3-20, 2002.
[15].
C. Callins Christiyana, V. Rajamani,
“Comparison of local binary pattern
variants for ultrasound kidney image
retrival,” IJARCSSE vol. 2, issue 10,
2012.
[16]. S. Manikandan, V. Rajamani, “Analysis
of ultrasound kidney image features for
image
retrival
by
gray
level
cooccurrence matrices” LNSE, vol. 1, no.
1, 2013.
[17]. J.
L.
Zhang,
H.
Rusinek,
H.
Chandarana, and V. S. Lee, “Functional
MRI of the kidneys,” J. Magn. Reson.
Imag. vol. 37, no. 2, pp. 282–293, 2013.
[18]. J. Tokuda, H. Mamata, R. R. Gill, N.
Hata, R. Kikinis, R. F. Padera, R. E.
Lenkinski, D. J. Sugarbaker, H. Hatabu,
“Impact of nonrigid motion correction
technique on pixel-wise pharmacokinetic
294
IJISET - International Journal of Innovative Science, Engineering & Technology, Vol. 2 Issue 4, April 2015.
www.ijiset.com
ISSN 2348 – 7968
analysis
of
free-breathing
pulmonary
dynamic contrast-enhanced MR Imaging, ”
J. Magn. Reson. Imag. vol. 33, no. 4,
pp. 968–973, 2011.
[19]. L. Bokacheva, H. Rusinek, J. L.
Zhang, Q. Chen, V. S. Lee, “Estimates of
Glomerular
Filtration
Rate
from
MR
Renography and Tracer Kinetic Models,”
J. Magn. Reson. Imag. 2011.
[20]. A.
M.
Khan,
Ravi
S.,
“Image
segmentation
methods:
a
comparative
study,” IJSCE, ISSN: 2231-2307, vol. 3,
issue 4, 2013.
[21]. E. Hodneland, E. A. Hanson, A.
Ludervold, J. Modersitzki, E. Eikefjord,
A. Z. Munthe-Kass, “Segmentation-driven
image registration application to 4D
DCE-MRI
recordings
of
the
moving
kidneys,” IEEE Trans. on Imag. Proces.
vol. 23, no. 5, 2014.
[22]. “Diagnosis and management of chronic
kidney disease” Scottish intercollegiate
guideline network, 2008.
[23]. J. S. Jose, R. Sivakami, N. U.
Maheshwari, R. Venkatesh, “An efficient
diagnosis
of
kidney
images
using
association rules,” IJCTEE, vol. 2,
issue 2, 2012.
[24]. Devidas
D.
Dighe,
“Edge
based
segmentation using SUSAN edge detector”
in national Conference “Intelligent Data
Analytics And Image Processing (IDAIP2007)
held at Bannari Amman Institute
of Technology, Tamilnadu Dt., 26-27 OCT
2007.
[25]. D. D. Dighe, J. J. Chopade, N. L.
Bhale, “Image segmentation using SUSAN
edge detector” In IEEE international
conference “SPIT-IEEE Colloquium 2007 and
International Conference” held at Sardar
Patel Institute of Technology Mumbai on
4-5 Feb 2008.
Nikkita U Derle has completed B. E. Electronics and
Telecommunication from Savitribai Phule Pune University and pursuing
M. E. E&TC (VLSI and Embedded System) from Matoshri College of
Engineering and Research Centre, Nasik-105. Her field of interest are
signal processing, image processing.
Devidas D. Dighe has completed BE Electronics from Savitribai Phule
Pune University and ME Electronics (Digital Systems) from Government
College of Engineering, Pune, currently pursuing PhD from Amravati
University. His major field of studies are Digital Systems, Signal
Processing, Image processing. He is working as Associate Professor and
Head in E&TC at Matoshri College of Engineering and Research Centre,
Nashik-105.
295