Abstract.

Keywords: MRI, brain tumor segmentation, Image processing, preprocessing

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

1            
Introduction

Brain tumor is one of the dangerous disease
which is curable only if diagnosed at early stage. According to SEER, in 2014,
total of 162,341 persons were suffering from brain tumor and other nervous
system cancers in US. There are approximately 23,680 new cases of brain tumor
patients in year 2017. And around 70% of these people may die because of this
cancer, if not diagnosed on time. There exist different imaging techniques to
capture brain tissue images like X-ray, CT scan i.e., PET scan, MEG, MRI etc.
But among these techniques MRI i.e. Magnetic Resonance Imaging is the most
popular technique 1. Advantages of MRI are discussed in subsection
1.3.

Manual segmentation of these MRI images takes
time for actual diagnosis. Also, the results of segmentation are most of the
times subjective as every other clinical expert may have different opinion. In
contrast to this, semi-automatic and automatic segmentation of brain MRI reduces
the diagnosis time, so that clinical expert can implement required treatment as
early as possible. This paper gives an overview of segmentation methods and
current trends in segmentation of MRI.

In the following subsections, principle of
MRI and Advantages of MRI are discussed in brief. Also, concept of brain tumor
segmentation from biology perspective and image processing perspective is
discussed in subsection 1.3. In section 2, overview of available brain MRI
databases and different imaging modalities are presented. Section 3, deals with
various preprocessing techniques. In section 4, various brain MRI segmentation
methods are discussed. Different validation techniques for brain MRI
segmentation are discussed in section 5. And section 6 gives possible future
trends in MRI segmentation.

 

1.1        
Principle of MRI

Human body contains mostly water
which has Hydrogen nuclei. MRI scanner applies strong magnetic field in the
range varying from 0.2 to 3 teslas. This magnetic field aligns the hydrogen
nuclei proton spins.

When magnetic field is applied, protons
absorb energy and flip their spins. When the field is turned off, the protons
gradually come back to their normal spins. This phenomenon is called
precession.

This precession process produces a radio
signal, which is measured by receivers and made into image. So, final image
produced is result of the precession.

1.2        
Advantages and uses of MRI

As seen from section 1.1, we can see that
MRI does not involve any type of radiation. So, it is considered safe over
X-ray imaging and CT scan. Also, MRI is non-invasive method of brain imaging.

The complete process of taking MRI is
simple from patient’s point of view. He or she does not suffer through any
pains while undergoing MR imaging. MRI gives good soft tissue contrast.

MRI is used for diagnosis of brain tumor as
well as diseases like spinal cord injuries, aneurysms, swellings, stroke,
infection etc. But, due to above mentioned advantages of MRI, it has become
most popular diagnosis tool for brain tumor in the last 15-20 years.

1.3        
Concept of brain MRI
segmentation

Segmentation is a process of delineating a
region into different parts. In brain tumor segmentation, tumor area is
annotated separately from normal brain tissues. Gray matter(GM), White matter(WM)
and Cerebrospinal fluid(CSF) together constitute healthy brain tissues.
Whereas, Edema, necrotic core and active region are parts of tumor tissues.
These tissues are shown in figure 1a and 1b.

  

  

(a)   
                                                          (b)

Fig.
1.a : Healthy brain tissues, b:
Tumor tissues

From image processing point of view, brain
MRI segmentation process can be divided into blocks shown in figure 2. Image
acquisition step consists of acquiring MRI images for analysis. Preprocessing
of these images is an essential step so as to have correct tumor segmentation.
Finally, validation of segmentation is necessary to check if the segmentation
technique implemented is practically viable.

Fig.
2. Overview of brain MRI segmentation using image processing

2            
Brain MRI image acquisition

The researcher can collect the MRI images
from hospitals or he/she can use available databases. There are different MRI
modalities which are as follows 2-

 1. 
T1-weighted: This shows healthy
brain tissues easily

 2. 
T1-weighted with contrast
enhancement: Brain tumor borders are seen brighter

 3. 
T2-weighted: The edema is shown
brighter

 4. 
Proton density-weighted:

 5. 
Fluid attenuated inversion
recovery MRI: Edema is effectively separated from CSF region

 In
the literature studied, we find following databases are popularly used-

·   
BRaTS

·   
Brainweb

·   
IBSR

3            
Preprocessing of brain MRI

Before doing actual segmentation of brain tumor,
the MRI images should be treated with preprocessing operations. Basic
preprocessing operations are image denoising, removing non-cerebral tissues and
intensity normalization.

Noise in MRI follows a Rician distribution.
Unlike Gaussian noise, which is additive in nature, Rician noise depends on
signal. Rician noise introduces signal dependent bias which reduces image
contrast 3. And, the segmentation process becomes difficult.

Wavelet transform is popularly used for
removing the Rician noise from MRI. Kinita et al used wavelet transform for
denoising MRI. The input MRI is decomposed into wavelet coefficients. These
coefficients are then thresholded considering properties of Rician noise. And,
the MRI is reconstructed using these thresholded coefficients 4. Technique of
nonlocal means is implemented for denoising in 5. This method gives good
result in denoising of MRI with high signal-to-noise ratio(SNR). This filtering
technique looks for redundancy in image in order to remove noise. Anisotropic
diffusion filter also helps in removal of noise. Wave atom transform for
denoising MRI is proposed by Rajeesh et el 6. Figure 3 shows results of
Denoising on T1 images using wave atom transform.

Fig.
3. Result of denoising MRI using wave atom transform 6

Skull stripping involves removal of
non-cerebral tissues. Non-cerebral tissues include skull, scalp and meninges.
This preprocessing technique lowers the chances of misclassification. Accurate
skull stripping is a challenging process as tumor may be attached to skull in
some cases. A detailed overview of current trends in skull removal is presented
in this paper 7. Sudipta et. Al. proposed convex hull algorithm to create a
brain mask 8. All the pixels which are inside the convex hull are set with
value 1 and those which are lying outside convex hull are given value 0, which
eventually results in removal of skull. William Speier et al proposed an
extended ROBEX skull stripping method 9. Adaptive thresholding is applied on
brain boundaries, to find potential resection space or cavity. And, the leakage
of the ventricles is corrected by random walker method.

Intensity Normalization is a process of
normalizing various MRI modalities onto one single range. This step is useful
when clustering methods are to be used for segmentation.

4            
Segmentation Methods

Brain tumor is segmented from brain MRI
using different segmentation algorithms based on various features of input MRI
image. Various features like mean, entropy, energy, variance, skewness,
contrast are extracted from MRI image, which forms basis for tumor segmentation
10.

5            
Validation of segmentation
results

6            
Conclusion

7            
References

  
1.   P. Y. Wen, D. R. Macdonald,
D. A. Reardon, T. F. Cloughesy, A. G. Sorensen, E. Galanis, J. DeGroot, W.
Wick, M. R. Gilbert, A. B. Lassman, et al., Updated response assessment
criteria for high-grade gliomas: Response assessment in neuro-oncology working
group, Journal
of Clinical Oncology, vol. 28, no. 11, pp. 1963-1972, 2010.

   2.   Jin Liu, Min Li, Jianxin
Wang_, Fangxiang Wu, Tianming Liu, and Yi Pan: A Survey of MRI-Based Brain
Tumor Segmentation Methods, Tsinghua Science and Technology, ISSN
1007-0214,04/10, pp578-595, volume 19, Number 6, December 2014

  
3.   Nowak, R. D. “Wavelet-Based Rician Noise Removal for Magnetic
Resonance Imaging.” IEEE Transactions on Image Processing, vol. 8, no.
10, Oct. 1999, pp. 1408–19. IEEE Xplore, doi:10.1109/83.791966.

   4.   Kinita b vandara, Mr. N. R.
Patel, prof. H. H. Wandra, dr. H n pandya, mr. Vinod thumar

  Removing of Rician noise using
wavelet in magnetic resonance images, Journal of Information, Knowledge and Research in Electronics and Communication
engineering, ISSN: 0975 – 6779, nov 10 to oct 11, volume – 01, issue – 02

  
5.   Nocolas Wiest-Daessle, Sylvain Prima, Pierrick Coupe, Sean Patrick
Morrissey and Christian Barillot, Rician
Noise Removal by Non-Local means filtering for low Signal-to-Noise ratio MRI:
Applications to DT-MRI, MICCAI 2008, Part II, LNCS 5242, pp 171-179,
Springer-Verlag Berlin Heidelberg, 2008

  
6.   Rajeesh J, Moni RS, Kumar SP, GopalaKrishnan T (2010) Rician noise removal on MRI using wave atom
transform with histogram based noise variance estimation. In: 2010 International
Conference on Communication Control and Computing Technologies. pp 531–535.

  
7.   Kalavathi, P., and V. B. Surya Prasath. “Methods on Skull Stripping
of MRI Head Scan Images—a Review.” Journal of Digital Imaging, vol. 29,
no. 3, June 2016, pp. 365–79. PubMed Central, doi:10.1007/s10278-015-9847-8.

   8.   Sudipta Roy, Sanjay Nag, Indra Kanta Maitra, Prof. Samir Kumar
Bandyopadhyay, Artefact Removal and Skull Elimination from MRI of Brain Image,
International Journal of Scientific & Engineering Research, Volume 4,
Issue 6, June-2013 163 ISSN 2229-5518

  
9.   Speier, William, et al. “Robust Skull Stripping of Clinical
Glioblastoma Multiforme Data.” Medical Image Computing and Computer-Assisted
Intervention – MICCAI 2011, Springer, Berlin, Heidelberg, 2011, pp. 659–66.
link.springer.com, doi:10.1007/978-3-642-23626-6_81.

10.  
Kharat, K. D., Pawar, V. J.,
& Pardeshi, S. R. (2016). Feature extraction and selection from MRI images
for the brain tumor classification. In 2016 International Conference on
Communication and Electronics Systems (ICCES) (pp. 1–5). https://doi.org/10.1109/CESYS.2016.7889969

Author