mri image brain tumor detection and segmentation · segmentation of mri images is challenging due...

MRI Image Brain Tumor Detection and Segmentation

P. Ratha1 and Dr. B. Mukunthan2 1. Assistant Professor, Department of Computer Science, Bharathidasan University Model College, Aranthangi.

E-mail: [email protected]

2. Associate Professor, Department of Computer Science, Sri Ramakrishna College of Arts & Science,

Coimbatore. E-mail: [email protected]

Abstract - Detection of brain tumor is very common fatality in current scenario of health care society. Image

segmentation is used to extract the abnormal tumor portion in brain. Brain tumor is an abnormal mass of tissue

in which cells grow and multiply uncontrollably, apparently unregulated by mechanisms that control cells.

Segmentation of brain tissue in the magnetic resonance image (MRI) is very important for detecting and

existence of outlines the brain tumor. In this research an algorithm for segmentation based on the symmetry

character of brain image is presented. Our goal is to detect the position and edge of tumors automatically.

Restorative picture dissection What's more preparing need extraordinary importance in the field about medicine,

particularly clinched alongside non-invasive medicine and clinical study. It aides those doctors to visualize

Furthermore examine the picture to see abnormalities previously, internal structures. This suggested system

comprises for four phases. Over 1st phase MRI picture is procured toward utilizing MATLAB. In the second

stage pre-processing need been done. This pre-processed MRI cerebrum picture is normalized also improved to

attain computational consistency. In the third stage high back commotion parts are evacuated toward

suitableness filters. For fourth stage the tumor a piece need been fragmented utilizing successful hereditary

calculation and the execution Investigation need been made.

Keywords - MRI, Segmentation and Detection, Bio-Medical, Morphological operations.

I.INTRODUCTION

The field known as biomedical analysis has evolved considerably over the last couple of decades. The

widespread availability of suitable detectors has aided the rapid development of new technologies for the

monitoring and diagnosis, as well as treatment, of patients. Over the last century technology has advanced from

the discovery of x-rays to a variety of imaging tools such as magnetic resonance imaging, computed tomography,

positron emission tomography and ultrasonography. The recent revolution in medical imaging resulting from

techniques such as computed tomography (CT) and magnetic resonance imaging (MRI) can provide detailed

information about disease and can identify many pathologic conditions, giving an accurate diagnosis.

Furthermore, new techniques are helping to advance fundamental biomedical research. Medical imaging is an

essential tool for improving the diagnoses, understanding and treatment of a large variety of diseases [1, 7, 8].

The extra-ordinary growth experimented by the medical image processing field in the last years, has motivated

the development of many algorithms and software packages for image processing.

Segmentation of MRI images is challenging due to poor image contrast and artifacts that result in missing or

diffuse tissue boundaries. We present a discrete wavelet based genetic algorithm is proposed to detect the MR

brain Images. First, MR images are enhanced using discrete wavelet descriptor, and then genetic algorithm is

applied to detect the tumor pixels. A genetic algorithm is then used in order to determine the best combination

of information extracted by the selected criterion.

JASC: Journal of Applied Science and Computations

Volume 5, Issue 9, September/2018

ISSN NO: 1076-5131

Page No:644

The present approach uses k-Means unsupervised clustering methods into Genetic Algorithms for guiding this

last Evolutionary Algorithm in his search for finding the optimal or sub-optimal data partition (Harris and

Buxton 1996, Kim et al 2000, Li Zhijun et al 2006) task that as we know, and requires a non-trivial search

because of its intrinsic NP-complete nature. To solve this task, the appropriate genetic coding is also discussed

since this is a key aspect in the implementation. Our purpose is to demonstrate the efficiency of Genetic

Algorithms to automatic and unsupervised image segmentation. Some examples in human MRI brain tumor

segmentation are presented and overall results discussed.

Medical imaging is performed in various modalities, such as magnetic resonance imaging (MRI), computed

tomography (CT), ultrasound etc. Segmentation is typically performed manually by expert physicians as a part of

treatment planning and diagnosis. Due to the increasing amount of available data and the complexity of features

of interest, it is becoming essential to develop automated segmentation methods to assist and speed-up image

understanding tasks.

II. RELATED WORK

Image segmentation is a low-level image processing task that aims at partitioning an image into

homogeneous regions. How region homogeneity is defined depends on the application. A great number of

segmentation methods are available in the literature to segment images according to various criteria such as grey

level, color, or texture (Gonzales and woods 2002). Several automated methods have been developed to process

the acquired images and identify features of interest, including intensity-based methods, region-growing

methods and deformable contour models. Intensity-based methods identify local features such as edges and

texture in order to extract regions of interest. Region-growing methods start from a seed-point on the image and

perform the segmentation task by clustering neighborhood pixels using a similarity criterion. Recently,

researchers have investigated the application of genetic algorithms into the image segmentation problem

(Nordin and Banzhaf 1996, Peng-Yeng 1999 and Ou et al 2004).

To improve the image quality we can use any one of the filtering technique (Mostafa et al 2001). Magnetic

Resonance (MR) image enhancement are mainly used for reconstruction of missing or corrupted parts of MR

images, image de-noising and image resolution enhancement. While using Magnetic Resonance (MR) images

resolution enhancement face many problems like Resolution enhancement of MR images (512 x 512 pixels 2

times more),conservation of sharp edges in the image and conservation and highlighting of details. There are

two designed and tested methods used for image resolution enhancement: Discrete Fourier Transform (DFT)

and Discrete Wavelet Transform (DWT). Recently wavelets have been successfully used in a large number of

biomedical applications (Mostafa et al 2001 and Bealy 1992). The multi-resolution framework makes wavelets

into very powerful compression and filter tool and the time and frequency localization of wavelets makes it into

a powerful tool for feature detection. This chapter, 2D discrete wavelet transform is used for removing noise

from MRI brain image.

A genetic algorithm is based on the idea that natural evolution is a search process that optimizes the structures it

generates. An interesting characteristic of GA is their high efficiency for difficult search problems without being

stuck in local extreme. In a GA, a population of individuals, described by some chromosomes, is iteratively

updated by applying operators of selection, mutation and crossover to solve the problem. Each individual is

evaluated by a fitness function that controls the population evolution in order to optimize it. GA can be used to

find out the optimal label of each pixel, to determine the optimal parameters of a segmentation method, or to

merge regions of a fine segmentation result.



ISSN NO: 1076-5131

Page No:645

Concerning the fitness function, it can be an unsupervised quantitative measure of a segmentation result or a

supervised one using some a priori knowledge. In this chapter, we deal with a general scheme for MRI brain

tumor image segmentation that involves a GA. GA is used here as an optimization method for the optimal

combination of segmentation results whose quality is quantified through an evaluation criterion. We use a

general scheme to define segmentation methods by optimization.

III. PROPOSED WORK

This proposed method has been used for MR brain images which are affected by the brain tumor. the

input. Image enhancement is needed for further processing even though MR scan image having high contrast

compared with CT scan. The work is methods of image segmentation for extraction of tumor in the MRI

images. Practically it was working well for removing noises and the result is better than the filters is using the

genetic algorithm. The Basic approach of the comparison is to implement different segmentation of tumor area

of MRI images. Purposely this segmentation is intended to detect the tumor and the result is used to measure

the performance of the resultant image with various quality measurements.

3.1 METHODOLOGY

In this section, Dataset of brain MRI images is made and performs MRI segmentation algorithms.

Implementation process or steps of the system are explained in following section.

Fig1. Block Diagram of Proposed Work

A.Image Acquisition

The quality of every 256*256 slice acquired intra-operatively is fairly similar to images acquired with a 1.5

T conventional scanner, but the major drawback of the intra-operative image is that the slice remains thick (2.5

mm). Images of a patient obtained by MRI scan is displayed as an array of pixels (a two dimensional unit based

on the matrix size and the field of view) and stored in Mat lab7.0.Here, grayscale or intensity images are

displayed of default size 256 x 256.The following figure displayed a MRI brain image obtained in Mat lab 7.0.

The brain MR images are stored in the database in JPEG format. Fig 2 shows the image acquisition.

Start Read MRI brain

images Pre- processing

Tumor portion

Identify using

morphological

operators

De-noising

Ascertain range of

the tumor End



ISSN NO: 1076-5131

Page No:646

Fig.2 Image Acquisition

B. Pre-processing

Preprocessing functions involve those operations that are normally required prior to the main data

analysis and extraction of information, and are generally grouped as radiometric or geometric corrections.

Radiometric corrections include correcting the data for sensor irregularities and unwanted sensor or atmospheric

noise, removal of non-brain pixels and Converting the data so they accurately represent the reflected or emitted

radiation measured by the sensor .In this work tracking algorithm is implemented to remove film artifacts. The

high intensity value of film artifacts are removed from MRI brain image. During the removal of film artifacts,

the image consists of salt and pepper noise

Fig.3 Before and After Preprocessing

C.De-Noising

In a wide variety of image processing applications, it is necessary to smooth an image while preserving its

edges. The gray levels often overlap that makes any post-processing task such as segmentation, feature

extraction and labeling more difficult. Filtering is perhaps the most fundamental operation in many biomedical

image processing applications, where it reduces the noise level and improves the quality of the image. In general,

the problem of how to select a suitable de-noising algorithm is dependent on the specific targeted application.

(i) De-noising using Median Filter Median Filter can remove the noise, high frequency components from MRI without disturbing the edges

and it is used to reduce’ salt and pepper’ noise. This technique calculates the median of the surrounding pixels to

determine the new (de-noised) value of the pixel. A median is calculated by sorting all pixel values by their size,

then selecting the median value as the new value for the pixel. For each pixel, a 3*3, 5*5, 7*7, 9*9,

11*11windows of neighborhood pixels are extracted and the median value is calculated for that window. The

intensity value of the center pixel is replaced with the median value. This procedure is done for all the pixels in

the image to smoothen the edges of MRI. High Resolution Image was obtained when using 3*3 than 5*5 and so

on.



ISSN NO: 1076-5131

Page No:647

Fig4. 3*3,5*5 Median Filtered Images

D. Image segmentation Segmentation is process to extract information from complex medical images. The main objective of the image

segmentation is to partition an image into mutually exclusive and exhausted regions such that each region of

interest is spatially contiguous and the pixels within the region are homogeneous with respect to a predefined

criterion [6]. Single seed Region growing algorithm is used for segmentation for that threshold value and seed

point are needed.

(i) Single Seed Region Growing Seed point will be selected by the user for this segmentation. This single seed point is used to calculate

the neighbor pixel values. If the single seed point’s properties are similar to the neighborhood pixels, that are

added into one region. Otherwise, that is added into another region. Then calculate the x co-ordinates and y co-

ordinates of neighborhood pixels which are similar with the seed point. During region growing method can have

many regions but these regions are not similar with other. Stopping criteria should be efficient to discriminate

neighbor elements in non-homogeneous domain. There are 4 and 8-connected neighborhoods for adjacent pixel

relationship. In this system 8-connected neighborhood pixels adjacent relationship is used. In the region pixels

are added which have the nearest intensity to the mean of the region. Then the new mean of the region is

calculated. Finally it provides good segmented result.

E. Morphological operations

During segmentation, the image is converted into binary format based on the threshold value. Dilation

and erosion of the morphological operations are applied on the segmented image. The purpose of the

morphological operators is to separate the tumor part from the image. This portion has the highest intensity

than other regions of the image [1]. The dilation “grows” or “thickens” the objects in a binary image. The

Erosion “shrinks” or “thins” the objects in a binary image. In this morphological process, the commands

imitate, improve and steel are used. By using this morphological operations able to get tumor recognized image

[11].

F. Extract tumor position The result of the morphological operations is applied in the pre-processed image. The recognized image

is in the form of binary image. The white pixels are applied where is actually affected by the tumor and then

remove the gray matters. Finally, the skull is removed. It can be used to extract the tumor portion easily.

G. Tumor Area calculation The area is calculated with the total number of pixels present in the extracted tumor region. Calculate

size of the input image e.g., 555 * 740. The Horizontal resolution of the output image is 96 dpi. The Vertical

resolution of the output image is 96 dpi. From the horizontal and vertical resolution can find the area of a single

pixel.

Area of single pixel = (1/96) × (1/96) inch

Area of the tumor = Area of single pixel × total number of pixel (cm2) [6].



ISSN NO: 1076-5131

Page No:648

(i)False Positive Ratio (Over Segmentation)

It is the ratio of total number of pixels which are not actual pixels of segmented image to the total number of

pixels which are actual pixels of ground truth image.

(ii)False Negative Ratio (Low Segmentation)

It is the ratio of total number of pixels which are not actual pixels of ground truth image to the total number of

pixels which are actual pixels of ground truth image.

IV. RESULT AND DISCUSSION

The existing de-noising approach is modified for better segmentation. Implemented system shows the

results of two types of experiments. First, System uses soft segmentation based DWT for de noising MR images.

Shows the original MR image without de noising and the enhanced image for single level of decomposition.

Moreover, magnetic resonance images are lesser noise densities corrupted images, single level of DWT

decomposition is sufficient for this type of images. During the decomposition σ = 5,

SNR = 43.5 and σ = 50, SNR = 20.3, while σ value increases value of SNR will be reduced gradually.

The Table 1 is described that the result experimented by the improved de-noised approach with

segmentation. Many images are also tested with this novel approach. That shows improved de-noising method

can work well with more images, figure 5, fig (a), fig (b) and fig (c) denotes input grayscale image, enhanced

image of fig (a) and result of region growing segmentation of fig (b) are respectively. the results of

morphological operations are represented as fig (d), fig (e), and fig (e), that denotes eroded image with disk

shape structuring element and the size is 1, eroded image with disk shape structuring element and the size is 6 of

fig (d), dilated image with disk shape structuring element with the size is 6 of fig (e) are respectively.

Results of improved de-noise approach with segmentation

Description Input

image

Contrast

Adjusted and

de-noise

image

segmented

Image

Ascertain

range of the

tumor

Extracted

Tumor region

(a) (b) (c) (d) (e)

Figure 5.MRI impute image, de-noise image, segmented image and tumor extracted image.

This work is detects and extracts the location of the tumor portion, and calculate the area of the tumor

portion. The output of the segmented regions and the extracted tumor portion and its area is shown below. The

proposed system gives very reasonable results for different kind of MR Images.



ISSN NO: 1076-5131

Page No:649

Results for segmentation using existing novel de-noising.

Images Segmented,

De -noising method

Extracted Tumor

region

Image1

Image2

Image3

Image4

Figure 6: I1 to 4 in the 1st column are Input of MRI brain image, which is contains tumor. The results

of morphological operation and those results are applied into the gray input image. T1 to T10 in the 4th column

are the results of the tumor portion extracted from the 1 to 4 images.

Optimal threshold values, Number of pixels in the tumor portion and area of the tumor of the ten

different images are mentioned in the below table and its bar chart also shown in figure 6

Table 1.Area of Tumor position

MRI Images

Optimal

Threshold

Values

Number of

pixels in

tumour

Area of

Tumour

(cm^2)

imag1 0.5486 8773 0.7336

image2 0.575 19702 0.1126

image3 0.5686 24829 0.065

image4 0.4781 24891 0.0643

image5 0.7477 9389 0.0765



ISSN NO: 1076-5131

Page No:650

Figure 7

Present some genetic segmentation results of human MR brain tumor T1 and T2 weighted

(also sometimes called a genotype) is a set of parameters which define a proposed solution to the

the genetic algorithm is solve. The set of all sol

Quality Image Measurement The mean squared error (MSE) of an estimator measures the average of the squares of the "errors", that

is, the difference between the estimator and what is estimated. PSNR (Peak Signal

commonly used to measure the quality of reco

MSE and PSNR.

Table 2. Comparison between genetic algorithms with ground truth.

Patient ID

Grade level

397384 High

1941040 High

1953042 High

197906 Low

1956041 Low

1943061 Low

0

1

2

3

4

5

Figure 7: the chart for area calculation of tumor region.

resent some genetic segmentation results of human MR brain tumor T1 and T2 weighted

(also sometimes called a genotype) is a set of parameters which define a proposed solution to the

the genetic algorithm is solve. The set of all solutions is known as the population of images

The mean squared error (MSE) of an estimator measures the average of the squares of the "errors", that

is, the difference between the estimator and what is estimated. PSNR (Peak Signal

commonly used to measure the quality of reconstruction of an image. Figure 6 is drawn based on the values of

Comparison between genetic algorithms with ground truth.

level

Number of Detected Edges

MES PSNR

Segmentation

5259 4382 1997

5120 4323 1836

6807 5757 2302

1491 649 317

2509 1080 433

2567 1072 417

Optimal Threshold Values

Number of pixels in tumour

Area of Tumour (cm^2)

r region.

resent some genetic segmentation results of human MR brain tumor T1 and T2 weighted chromosome

(also sometimes called a genotype) is a set of parameters which define a proposed solution to the problem that

images.

The mean squared error (MSE) of an estimator measures the average of the squares of the "errors", that

is, the difference between the estimator and what is estimated. PSNR (Peak Signal to Noise Ratio) is most

is drawn based on the values of

Comparison between genetic algorithms with ground truth.

MES PSNR

1997

1836

2302

317

433

417

Optimal Threshold

Number of pixels in

Area of Tumour



ISSN NO: 1076-5131

Page No:651

Figure 8 is drawn based on the values of MSE and PSNR

Table

Patient ID

Lesion

397384 Left Frontal Parietal

19410407

Left High Parietal

19530428 Left Temporal Lobe

19790628 Left Frontal Parietal

19560416

Left Thalamus

19430618

Left High Parietal

0

1

2

3

4

5

6

Category 1 Category 2

0

1

2

3

4

5

6

Category 1

is drawn based on the values of MSE and PSNR

Table 3: Areas of tumor position:

Volume of

tumor areas

(Pixels)

% of Damage areas

Left Frontal Parietal 4315 17.26

Left High Parietal

1068

4.27

Left Temporal Lobe 435 1.74

Left Frontal Parietal 1776 7.10

Left Thalamus

1060

4.24

Left High Parietal

3824

15.30

Figure 8. Areas of tumor position

Category 2 Category 3 Category 4

MES

PSNR

segmantation range

Category 1 Category 2 Category 3 Category 4

position of tumour

volume of tumour

tumour position

is drawn based on the values of MSE and PSNR

% of Damage areas

segmantation range

position of tumour

volume of tumour

tumour position



ISSN NO: 1076-5131

Page No:652

The quality of images can also be measured by contrast and noisiness. The existing de-noised image is

not well worked with the segmentation and the statistical measurement is clearly. Figure 5 is a graphical

representation that shows the quality measurements of all the quality parameters improved de-noising method

results are compared with the original input image. Comparison of Tables 1 and 2 shows the changes of the

quality in the improved de-noise image which is better than the existing novel approach. Table 1 show that there

is a better improvement in the image quality than that Table 2. And this enhanced de-noising novel approach is

giving the fruitful results that are proved statistically and experimentally in Table 3.

V. CONCLUSION

In this work, it is mainly focused on segmentation and extraction of brain tumor. The single seeded region

threshold and morphological technique. The major decisions are choosing a method of segmentation to which

genetic algorithms will be applied, finding a fitness function that is a good measure of the quality of image

segmentation and finding a meaningful way to represent the chromosomes. We showed that this kind of

approach can be applied either for grey-level magnetic resonance images. The developed method uses the ability

of GA to solve optimization problems with a large search space (label of each pixel of an image). The developed

method can also integrate some a prior knowledge (such as a local ground truth) if it is available. The developed

method achieved SNR value from 20 to 44 and segmentation accuracy from 82 percent to 97 percent of

detected tumor pixels based on ground truth. The suggested novel approach of improved de-noising method

was equated with the being de-noising method. And the effects were employed the algorithm to find out the

dissimilarity between original and de-noise images. The experimental results establish the effectiveness of the

suggested work. In future, the tumor part is taken for classification of tumor types such as benign and

malignant.

REFERENCES [1] Anithadevi D, Perumal K. Brain tumor extraction based on segmentation. IJRITCC. 2014; 2(9):2682–9.

[2] Anithadevi D, Perumal K. Generic evaluation of automatic segmentation with seed selection. National Conference

Contemporary Developments in Information and Communication Technologies; Madurai. 2014. p. 14 24.

[3] Anithadevi D, Perumal K. Novel approach for noise removal of brain tumor MRI images. JBEMi. 2015; 2(3):1–14.

[5] Chang-Tsun Li and Randy Chiao (2003), ‘Multiresolution genetic clustering algorithm for texture segmentation’, Image and

VisionComputing, Vol. 21, pp. 955 - 966.

[6]. Kim E.Y., Park S.H. and Kim H.J. (2000), ‘A genetic algorithm-based segmentation of Markov random field modeled

images’, SignalProcessing Letters, IEEE, Digital Object Identifier10.1109/ 97.873564. Vol. 7, No. 11, pp. 301- 315.

[7]. Minglun Gong and Yee-Hong Yang (2001), ‘Genetic-based multiresolution color image segmentation’, Vision Interface, pp.

141–148.

[8]. Nordin P. and Banzhaf W. (1996), ‘Programmatic compression of images and sound’, In Genetic Programming

1996,Proceedings of the First

Annual Conference, Koza J. R. Editors, MIT Press, pp. 345-350.

[9] Ou Y., Cheng W. and Han Ferng-ching (2004), ‘Based on genetic algorithm fuzzy c- means clustering algorithm’, Journal of

Chongqing

University, Vol. 27 No. 6, pp. 89 - 926 . Philippe Andrey (1999), ‘Genetic algorithms applied to image segmentation’ Image and

Vision Computing, Vol. 17, pp.175- 187.

[10] M. S. Yang, Y. J. Hu, Karen, C. R. Lin and C. C. Lin, ―Segmentation techniques for tissue differentiation in MRI of

Ophthalmology using fuzzy clustering algorithms,‖Magnetic Resonance Imaging, vol. 20, issue 2, pp. 173-179, 2002.



ISSN NO: 1076-5131

Page No:653

[11] G. C. Karmakar and L. S. Dooley, ―A generic fuzzy rule based image segmentation algorithm,‖Pattern Recognition Letters,

vol. 23, issue 10, pp. 1215-1227, 2002.

[12] D. L. Pham and J. L. Prince, ―An adaptive fuzzy C-means algorithm for image segmentation in the presence of intensity

inhomogeneities,‖Pattern Recognition Letters, vol. 20, issue 1, pp. 57–68, 1999.

[13] Y. A. Tolias and S. M. Panas, ―On applying spatial constraints in fuzzy image clustering using a fuzzy rule-based

system,‖IEEE Signal Processing Letters, vol. 5, issue 10, pp. 245–247, 1998.

[14] Y. A. Tolias and S. M. Panas, ―Image segmentation by a fuzzy clustering algorithm using adaptive spatially constrained

membership functions,‖IEEE Transactions on Systems, Man, and Cybernetics - Part A: Systems and Humans, vol. 28, issue 3,

pp. 359–369, 1998.

[15] A. W. C. Liew, S. H. Leung, and W. H. Lau, ―Fuzzy image clustering incorporating spatial continuity,‖IEE Proceedings

- Vision, Image and Signal Processing, vol. 147, issue 2, pp. 185–192, 2000.

[16] D. L. Pham, ―Fuzzy clustering with spatial constraints,‖ in IEEE Proceedings International Conference on Image

Processing, New York, II-65–II-68, 2002.

[17] M. N. Ahmed, S. M. Yamany, N. Mohamed, A. A. Farag, T. Moriarty, ―A modified fuzzy C-means algorithm for bias

field estimation and segmentation of MRI data,‖IEEE Transactions on Medical Imaging, vol. 21, issue 3, pp. 193–199, 2002.

[18] S. C. Chen and D. Q. Zhang, ―Robust image segmentation using FCM with spatial constraints based on new kernel-

induced distance measure,‖IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), vol. 34, issue 4, pp.

1907-1916, 2004.

[19] L. Szilágyi, Z. Benyó, S. M. Szilágyii, H. S. Adam, and M. R. Brain ―Image segmentation using an enhanced fuzzy C-

means algorithm,‖ in 25th Annual International Conference of IEEE EMBS, pp. 17-21, 2003.

[20] M. N. Ahmed, S. M. Yamany, N. A. Mohamed, A. A. Farag, and T. Moriarty, ―Bias field estimation and adaptive

segmentation of MRI data using modified fuzzy C-means algorithm,‖IEEE Computer Society Conference on Computer Vision and

Pattern Recognition, vol. 1, pp. 250–255, 1999.



ISSN NO: 1076-5131

Page No:654

mri image brain tumor detection and segmentation · segmentation of mri images is challenging due...

Documents