Indian Journal of Science and Technology, Vol 9(44), DOI: 10.17485/ijst/2016/v9i44/97678, November 2016
ISSN (Print) : 0974-6846
ISSN (Online) : 0974-5645
Highly Efficient Segmentation and Classification of
Premature Infants Brain MR Images at Global and
Tissue Level
Tushar H. Jaware1*, K. B. Khanchandani1 and Anita Zurani2
SSGMCE, Near Krishna Cottage, District Buldana, Shegaon - 444203, Maharashtra, India; [email protected],
[email protected]
2
SGBAU, Near Tapovan Gate, Amravati - 444602, Maharashtra, India; [email protected]
1
Abstract
Objectives: Every year millions of babies born preterm and it is a serious issue in developed countries. Preterm birth is
worldwide problem of young children. Methods/Statistical analysis: Advances in Magnetic Resonance Imaging (MRI),
comprehensive images of the newborn brain is visualized non-invasively. Here we are focusing on the early assessment
of brain development in neonates and premature infants using multi stage segmentation and classification approach with
quantitative analysis. To achieve higher segmentation and classification accuracy neural network classifier are used. It
is also helpful in detection and classification of more number of brain tissues. Findings: The accurate segmentation and
classification of brain tissues is very much useful in evaluating the brain development of newborn and premature infants. Therefore, newborn brain MRI forms a crucial part in analytical neuro-imaging; principally in the neonatal stage.
Our method gives higher dice similarity index values and higher computational speed as compared to existing methods.
Application: Resulting segmentations are used for volumetric analysis and quantification of cortical descriptions. This
analysis plays vital role for neonatologists in early detection and diagnosis of neural impairments.
Keywords: Classification, MRI, Multi-kernel Support Vector Machine, Newborn, Preterm, Segmentation
1. Introduction
The premature infants born every year are around 15
million and rising day by day. This is more than 10 % in
births1. In last two decades, ratio of preterm births has
increased drastically. In developed and developing countries, premature birth is prime crisis of young children.
Annually more than 1 million young children die due to
the difficulties of premature birth. Babies with preterm
birth frequently face severe and lifetime health issues,
including cerebral palsy, vision loss, breathing problems, jaundice, and neurological impairments. As per
the National Centre for Health Statistics, United States
acquired a “C” on the 8th annual report with a preterm
birth rate of 9.6 % in 2014.
*Author for correspondence
The main reason of newborn infant’s death is preterm
birth, which is generally before 37 weeks of pregnancy.
Premature infants are frequently at a higher level of risk,
disability and health issues. World Health Organization
(WHO) has defined the stages of preterm birth:
Tremendously premature: < 28 weeks
Extremely premature: from 28 to < 32 weeks
Late premature: from 32 to < 37 weeks
MRI is non-invasive technique used for acquiring
the brain images of infants. MRI is a harmless method
to obtain the brain images of infants without ionized
radiations. Due to these advantages of MRI, paediatric
neuro-imaging has reformed the health care system. MRI
of the new born brain helps to identify abnormalities such
as hydrocephalus, congenital malformations, hypoxic
ischemic encephalopathy, infections and infarction3. Thus
Highly Efficient Segmentation and Classification of Premature Infants Brain MR Images at Global and Tissue Level
brain MRI forms a fundamental part in diagnostic neuro
radiology, particularly in the neonatal stage. MR brain
images are generally used for evaluation of brain development in preterm infants. This MR brain images provides
complete analysis based on surface, area, volume and
morphology, this supports to identify the complications
and risk issues in the children due to premature birth4.
Study of anatomy of developing brain is vital to depict
normal development and investigate factors that affect
brain growth. So quantitative neuroimaging studies using
MRI are increasingly being used to assess brain growth
and development in this vulnerable population5–7. In this
paper we are focusing on the early assessment of brain
development in neonates using multi stage segmentation
and classification approach with quantitative analysis.
Potential challenges in neonatal brain8–10 MR image
segmentation are as follows:
•
•
•
•
•
•
•
•
•
Low CNR(Contrast to Noise Ratio)
Intensity Inhomogeneity
Variable imaging parameters
Overlapping intensities
Partial volume effect
Non-intensity based borders
Shape Complexity
Susceptibility artifacts
Normal anatomical variations
The brain development in neonatal stage is crucial
and significant changes occur in this interval. The major
challenges to assess brain development of neonates are
size and structure variations of infants. In brain tissue
high level of water content is present, this causes overlapped intensity values and inconsistency among the
brain s­ tructures11–13.
2. Material and Methodology
The primary intention of my research is to design and
develop an approach for automatic segmentation and
classification of neonatal brain MRI.
Objectives
• Study and analysis of different variants of techniques
meant for improving performance in Neonates and
Premature Infants Brain segmentation process.
• Design and develop an efficient Neonates and
Premature Infants Brain segmentation approach with
the aid of supervised classification techniques.
2
Vol 9 (44) | November 2016 | www.indjst.org
The basic aim of proposed research work is to ­segment
brain at global and tissue level. At global level major
brain areas such as brainstem, cerebellum and two hemispheres are segmented. Gray Matter (GM), White Matter
(WM), and Cerebrospinal Fluid (CSF)14,15 are ­segmented
at tissue level.
In concern with newborn infants further segmentation of Gray matter as cortical and subcortical GM is
needed. Myelination is very important process and infant
continuously goes through myelination, so it is essential
for evaluation to segment myelinated and unmyelinated
white matter. Overall process of segmenting different
brain structures at different levels in newborn and premature infants is really exigent task.
From the approach16, two challenges are ­identified
to improve their approach further. The first one is the
inclusion of multi-kernel support vector machine
(MKSVM) instead of SVM in the second stage. The
inclusion of MKSVM could improve the classification
accuracy of SVM for both linear and non-linear dataset.
Here, radial basis function (RBF), polynomial function and quadratic will be used as kernel to improve
the SVM performance further. The second one is the
Levenberg-Marquardt based neural network taken into
consideration for classification instead of multiclass
k-nearest neighbor (MCNN) in the third stage. In KNN,
the main disadvantage is that it does not learn anything
from the training data and which can be slow if there
are a large number of training samples. Therefore, neural network based classifier is exploited to improve the
performance.
Proposed approach for segmentation of Brain MR
Images involves the following steps:
(1) Preprocessing,
(2) Region Segmentation.
T1-weighted (T1w) and T2-weighted (T2w) brain
MR images of a newborn are used as input for first
step of proposed algorithm. Then this MR image
goes through various pre-processing steps to improve
the quality of images which is beneficial for efficient
segmentation. The various steps involved in pre-processing are de-noising, affine registration, intensity
homogeneity correction and alignment to the radiological orientation. Gaussian filter is used for noise
removal.
In the second step, the region of interest (ROC) will be
extracted by using three stages.
Indian Journal of Science and Technology
Tushar H. Jaware, K. B. Khanchandani and Anita Zurani
Stage I: K
-Nearest Neighbor (KNN) will be used for assign
for voxels to one of the tissue classes are labelled.
Stage II: D
edicated analysis of the remaining voxels will
be performed using multi-kernel support vector
machine (MKSVM).
Stage III: F
inally, the segmented region will be identified using Levenberg-Marquardt based neural
­network.
Finally, various brain MR images will be ­collected
from benchmark database17 and subjected to developed technique for performance evaluation of
segmentation. MATLAB is used as development tool for
­implementation.
3. Results and Tables
Figure 1. Graphical representation of dice similarity
index.
Table 1 gives us the quantitative analysis of brain MR
images for different subjects. For this analysis we used
dataset of 30 images. Various parameters are calculated
like precision, dice similarity index, computational speed.
The result shows the improvement in these parameters as
compared to existing methods18–21.
Figure 1 shows the graphical representation of dice
similarity index. The higher DC value indicates good
segmentation accuracy. The matching between manual segmented images and our segmented image is given by DC.
The Figure 2 shows the different operations performed
on input brain MR images. Figure 2 (a) shows the T1 w&
T2w input Brain MR Images. (b) Shows the pre-processed image which is obtained using Gaussian filtering.
(c) Shows the contouring of brain portion by using this
brain portion is extracted and skull portion is removed.
(d) Shows the segmentation of different brain tissues and
finally (e) shows the classification of brain tissues using
multi classification approach.
Table 1. Quantitative analysis
Subject Precision Overlap Metric Recall
DC
Time
1
0.6181
0.764
1
0.764
8.803
4
0.6028
0.7522
1
0.752
9.892
5
0.6099
0.7577
1
0.757
9.45
11
0.8945
0.9443
1
0.944
10.37
12
0.8489
0.9183
1
0.918
10.29
14
0.8478
0.9177
1
0.917
9.442
Mean
0.737
0.8424
1
0.842
9.71
Vol 9 (44) | November 2016 | www.indjst.org
Figure 2. (a) Input T-1 w & T2-w MRI image. (b) Preprocessed images (c) Intracranial cavity contour (d)
Segmented output (e) Tissue classification.
Indian Journal of Science and Technology
3
Highly Efficient Segmentation and Classification of Premature Infants Brain MR Images at Global and Tissue Level
4. Conclusion
Accurate segmentation and classification of prime ­tissues
of brain like GM, WM, and CSF is presented using multistage classification technique. Resulting segmentations is
used for volumetric analysis and quantification of cortical
characteristics. This analysis plays vital role for neurologist
and neonatologist in early detection of neural disorders
and neural impairments. Proposed work will provide
imperative information to neurophysicians for diagnosis
and better treatment planning of newborn babies. The pre
and better treatment planning of neural impaired infants
will help society to decrease count of mentally challenged
infants.
5. References
1. Kolk AG, Hendrikse J, Zwanenburg JJM, Visser F, Luijten
PR.Clinical applications of 7T MRI in the brain. European
Journal of Radiology. 2013; 18(2):708–18.
2. Pennock JM.Pediatric brain MRI: applications in neonates
and infants. eMagRes; 2007.
3. Dubois J, Benders M, Borradori-Tolsa C, Cachia A,
Lazeyras F, Leuchter HVR, Sizonenko SV, Warfield SK,
ManginH. Primary cortical folding in the human newborn:
an early marker of later functional development. Brain.
2008; 131:2028–41.
4. Rathbone R, Counsell SJ, Kapellou O, Dyet L, Kennea N,
Hajnal J, Allsop JM, CowanE. Perinatal cortical growth and
childhood neuro cognitive abilities.Journal of Neurology.
2011; 77(16):1510–17.
5. Osareh A, Shadgar B. A computer aided diagnosis system for breast cancer. International Journal of Computer
Science. 2011; 8(2).
6. Acır N, Ozdamar O,Guzelis C. Automatic classification of auditory brainstem responses using SVM-based
features election algorithm for threshold detection.
Engineering Applications of Artificial Intelligence. 2006;
19:209–18.
7. Valentini G, Muselli M,Ruffino F. Cancer recognition with
bagged ensembles of support vector machines. Neuro
Computing.2004; 56:461–6.
8. Zhang YL, Guo N,Du H,Li WH. Automated defect recognition of C-SAM images in IC packaging using support
4
Vol 9 (44) | November 2016 | www.indjst.org
vector machines. The International Journal of Advanced
Manufacturing Technology.2005; 25:1191–6.
9. Karabatak M, Ince MC. An expert system for detection of
breast cancer based on association rules and neural network. Expert Systems with Applications. 2009; 36:3465–9.
10. Fatih AM. Support vector machines combined with feature
selection for breast cancer diagnosis. Expert Systems with
Applications. 2009; 36:3240–7.
11. Freeman WT, Roth M. Orientation histograms for hand gesture recognition. In International Workshop on Automatic
Face-and Gesture-Recognition; 1995.p.296–301.
12. Sharma M,Dubey RB, Sujata, Gupta SK. Feature extraction of mammogram. International Journal of Advanced
Computer Research.2012; 2(5).
13. Singh S, Gupta PR. Breast cancer detection and classification of histopathological images. International Journal of
Engineering Science and Technology.2011; 3(5).
14. Priya VV, Shobarani. An efficient segmentation approach
for brain tumor detection in MRI.Indian Journal of Science
and Technology. 2016 May; 9(19).
15. Sasirekh N, Kashwan KR. Improved segmentation of MRI
brain images by denoising and contrast enhancement.
Indian Journal of Science and Technology. 2015 Sep; 8(22).
16. Moeskops P, Benders MJNL,Chiţa SM, Kersbergen KJ,
Groenendaal F,de Vries LS, Viergever MA, Isguma I.
Automatic segmentation of MR brain images of preterm
infants using supervised classification.Journal of Neuro
Image.2015;118:628–41.
17. Available from: http://mrbrains13.isi.uu.nl/.
18. Kharat KD, Kulkarni PP, Nagori MB. Brain tumor classification using neural network based methods.International
Journal of Computer Science and Informatics. 2012; 1(4).
19. Ali SM, Abood LK, Abdoon RS. Brain tumor extraction
in MRI images using clustering and morphological operations techniques. International journal of Geographical
Information System Application and Remote Sensing. 2013;
4(1).
20. Cheng I, Miller SP, Duerden EG,Sun K, Chau V, Adams
E, Poskitt KJ, Branson HM, Basu A. Stochastic process for
white matter injury detection in preterm neonates.Journal
of Neuro- Image.2015; 7:622–63.
21. Makropoulos, Gousias, Ledig, Aljabar,Serag,Hajnal,
Edwards, Counsell ,Rueckert. Automatic whole brain
MRI segmentation of the developing neonatal brain.IEEE
Transactions onMedical Imaging. 2014; 33(9):1818–31.
Indian Journal of Science and Technology