Advanced Morphological Technique for Automatic Brain Tumor ...

Available online at www.sciencedirect.com

ScienceDirect Procedia Technology 24 (2016) 1374 – 1387

International Conference on Emerging Trends in Engineering, Science and Technology (ICETEST - 2015)

Advanced morphological technique for automatic brain tumor detection and evaluation of statistical parameters K. SUDHARANIa,*, Dr.T.C. SARMAb, Dr. K. SATYA PRASADc a

Associate Professor, VNR Vignana Jyothi IET, Hyderabad, Andhra Pradesh, b Former Deputy Director, NRSA , Hyderabad, Andhra Pradesh, c Professor, JNTU Kakinada, Kakinada, Andhra Pradesh ,

Abstract A tumor is uncontrolled growth of the abnormal tissue in the body. If this phenomenon is in brain is brain tumor. A tumor may lead to cancer. Image processing techniques are applied to magnetic resonance (MR) images to detect tumor and edema. The main objective of this paper is to present the automatic segmentation method which separates non-enhancing brain tumors from healthy tissues in MR images by locating tumor position in the brain and to give complete statistical analysis of the tumor. By applying the algorithm presented in this paper we can determine the area of the tumor in the brain along with the area length in the vertical and horizontal planes, sensitivity of the tumor, specificity, similarity index can also be find. The knowledge of this information regarding tumor in the brain is important for diagnosis, planning, and treatment. The proposed algorithm can also be applied to the ground truth images. With the proposed algorithm the tumor detection and localization system was found to be able to accurately detect and localize brain tumor and processing time is less. This will help the physicians in analysing the brain tumors accurately and efficiently. © 2016 2016The TheAuthors. Authors.Published Published Elsevier © by by Elsevier Ltd.Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the organizing committee of ICETEST – 2015. Peer-review under responsibility of the organizing committee of ICETEST – 2015 Keywords: Detection and Localization System ; Area ; Sensitivity ; Specificity ; Similarity Index ; Ground Truth Images.

1. Introduction A brain tumor arises due to an abnormal growth of cells that have proliferated in an uncontrolled manner. Primary brain tumors can start from brain cells, the membranes around the brain (meninges), nerves, or glands.

* Corresponding author. Tel.: +0-000-000-0000 ; fax: +0-000-000-0000 . E-mail address: [email protected]

2212-0173 © 2016 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the organizing committee of ICETEST – 2015 doi:10.1016/j.protcy.2016.05.153

K. Sudharani et al. / Procedia Technology 24 (2016) 1374 – 1387

Tumors can directly destroy brain cells. They can also damage cells by producing swelling, increasing pressure within the skull. Image processing techniques are widely used to extract information from images. Image segmentation is one of them and image segmentation can be defined as the process of partitioning an image of any type into non-intersecting regions such that each region is homogeneous and represents a physical or conceptual entity. Image segmentation is a fundamental processing technique in many important medical applications. The objective of image segmentation is to partition an image into homogeneous regions and locate the contours of the regions [1, 2]. The aim and implementation of segmentation is very much dependent on data types and applications. Magnetic resonance imaging (MRI) is one of the most popular techniques for human brain imaging due to its versatility and flexibility. As a tumor consists of different biologic tissues, one type of MRI cannot give complete information about abnormal tissues. Combining different complementary information can enhance the segmentation of the tumors. The MR feature images used for the segmentation consist of three weighted images (T1, T2 and proton denqasity (PD)) for each axial slice through the head. The MR segmentation methods have been quite successful and are still very much in the development stages for pathological tissues some success recorded for specific disease processes [3,4,5]. Automatic brain tumor segmentation from MR images is a difficult task that involves various disciplines. There are many issues and challenges associated with brain tumor segmentation. Brain tumors may be of any size, may have a variety of shapes, may appear at any location, and may appear in different image intensities. Some tumors also deform other structures and appear together with edema that changes intensity properties of the nearby region. For many human experts, manual segmentation is a difficult and time consuming task, which makes an automated brain tumor segmentation method desirable. There are many possible applications of an automated method, it can be used for surgical planning, treatment planning, and vascular analysis [6]. Many automatic segmentation methods have been applying to the images to get desired results. Many researchers have proposed different methods to detect brain tumor, they have their own approaches. According to, CT-Scan technique usually used for monitoring the images of damaged brain part. The images of the CT Scans are shown in the form of gray scale images as the equipment for CT scans support this form of image color and for easy detection of tumor from the image [7]. MRI images give accurate information than CT images when it comes to soft tissues. The method of [8] uses Eigen image analysis which nicely shows tumor segmentation on images, though operator intervention is required to select a region of interest (ROI). Knowledge based techniques with some clustering method were used to enhance the effected part in the brain [9, 10]. Image Thresholding is the most popular segmentation method due to its instinctive properties and simple implementation. Thresholding selected for efficient segmentation results. Automated segmentation methods [11, 20]based on artificial intelligence techniques were proposed a technique to detect tumors from MR images using fuzzy clustering technique. This algorithm uses fuzzy C-means but the major drawback of this algorithm is more processing time required. In [12] carried out outlier detection followed by geometric and spatial constraints. The reference template image limits the efficiency of the results to the accuracy of the template and so makes it non-real time. Authors in [13] employed the technique of symmetry integration in several steps associated with segmentation, clustering and classification. However, use of small and unstructured dataset restricts the generality and clinical applicability. None of the methodologies mentioned above does not give mathematical information regarding its area, length, specificity, and other. In [14] mentioned about the area but they used histogram and Thresholding techniques only to detect the tumor and to calculate the area the drawback is sometimes manual Thresholding is required when the tumor in the image is low intensity and at that time the proposed algorithm may not be able to detect the tumor. The proposed method is able to detect very low intensity parts of the image also. 2.

Methodology

The proposed methodology includes several steps those are

1375

1376

K. Sudharani et al. / Procedia Technology 24 (2016) 1374 – 1387

x x x x x x x x x x

adjusting the brightness, contrast of the original image Resampling the image Color plane extraction Application of histogram techniques Determine measurements of the required portion Apply Thresholding to the previous step image so that only tumor region gets enhanced Fast Fourier transformation Histogram application for second time to get equalization of all desired pixels Advanced morphological techniques to get detection of tumor region Through practical analysis of the tumor detected image area length and other parameters will be obtained.

1377

K. Sudharani et al. / Procedia Technology 24 (2016) 1374 – 1387

Fig.1 Flow chart of the proposed algorithm These are the image processing techniques which are applied to detect and localize the tumor. 2.1. Brightness Adjustment Image brightness (or luminous brightness) is a measure of intensity after the image has been acquired. The proper brightness levels are of the utmost importance when processing an image. Contrast is the visual property of an object that separates it from other objects in an image. The contrast of objects against the background of an image is important for two functions: identifying an object and later tracking it. Gamma is used to quantify contrast and keep images “linear” which allows for objects to be defined more easily and generally made more uniform and therefore easier to track [15]. The mathematical expressions are: Brightness adjustment: S(x, y) = R(x, y) + k (1) Contrast adjustment: S(x, y) = R(x, y) + k (2) Power law transformation (γ): S(x, y) = [R(x, y)]γ=1 (3) Where S(x, y) is result of R(x, y) and R(x, y) original image and S(x, y) is the result of corresponding technique. 2.2. Re-Sampling of the Image Changing the dimensions of the image like width, height and resolution through bilinear interpolation. Bilinear Interpolation is a resampling method that uses the distance weighted average of the four nearest pixel values to estimate a new pixel value. Height * width =250*250 pixels.

(4)

2.3. Color Plane Extraction As the proposed method is method require accurate images each step will be very important in many aspects as the image is in grey scale image. To make suitable for the further steps 32-bit image will be converted to 8-bit image. We need to remove intensity plane from the image obtained in the last step. Color plane extraction = (image obtained in the last step) – (intensity plane) (5) 2.4. Histogram Processing •

An Image Histogram is a type of histogram that acts as a graphical representation of the lightness/color distribution in a digital image. It plots the number of pixels for each value.

•

It is common practice to normalize a histogram by dividing each of its values by the total number of pixels in the image, denoted by n. Thus,

•

H(n) Where n= 0, 1,… L -1.

(6)

1378

K. Sudharani et al. / Procedia Technology 24 (2016) 1374 – 1387

•

Thus,H(n)gives an estimate of the probability of occurrence of gray level. Note that the sum of all components of a normalized histogram is equal to 1.

• •

In order to smooth the histogram H(n), local weighted averaging technique is applied over the histogram Information derived from histograms are quite useful in other image processing applications, such as image compression and segmentation.

2.5. Tumor Measurements In this process all the geometrical measurements are taken. The geometrical measurements like area of the tumor part and length of tumor part in vertical and horizontal sections. All these measurements can be directly calculated in software itself. All measurements are in pixels. Later these pixels are converted into mm for convenient. So surgeon can easily understand about tumor and due to this process explanation is also very easy. 2.6. Thresholding Thresholding makes it possible to highlight pixels in an image. Thresholding can be applied to gray scale images or color images. In this discussion gray scale images are used. In Thresholding a pixel intensity value is adjusted, by taking the given value as reference the low intensity pixels will become zero and rest of the pixels will become 1. The result of the Thresholding is a binary image containing black and white pixels [17].

(7)

D(x, y) is result image after applying Thresholding, J(x, y) is the image from the previous process and K is any constant intensity value. 2.7 FFT Fourier Transform decomposes an image into its real and imaginary components which is a representation of the image in the frequency domain. If the input signal is an image then the number of frequencies in the frequency domain is equal to the number of pixels in the image or spatial domain. The inverse transform re-transforms the frequencies to the image in the spatial domain presented in [18]. The FFT of a 2D image is given by the following equations: (8)

Where f(n,m) is the pixel at coordinates (n,m), F(x,y) is the value of the image in the frequency domain corresponding to the coordinates x and y, N and M are the dimensions of the image. The dimensions of the image are a power of two. 2.8 Lookup Table A lookup table transformation involves conversion of input image gray level values to other gray level values in transformed image.

K. Sudharani et al. / Procedia Technology 24 (2016) 1374 – 1387

1379

T(x)= dynamicMin if x ≤ rangeMin f(x) if rangeMin < x ≤ rangeMax dynamicMax if x> rangeMax x represents the input image gray-level value dynamicMin = the smallest initial pixel dynamicMax = the largest initial pixel value dynamicRange=dynamicMax – dynamicMin f(x) represents the new value of pixel in the transformed image. 2.9 Advanced Morphology Morphology is a technique for extracting the information from an image which is representation and description of region shape. In this paper morphological operations are used in post processing mainly as a filter. Its fundamental operations are Boundary pixels and low frequency pixels are eliminated from image. Erosion: it shrinks objects in the binary image Dilation: grows or thickens the objects in binary image 2.10. Practical Analysis The complete analysis of the final image, desired result, will be displayed in a spread sheet.

3.

Proposed algorithm results and their mathematical calculations

1380

K. Sudharani et al. / Procedia Technology 24 (2016) 1374 – 1387

Fig.2 MRI Images applied to the proposed algorithm

1381

K. Sudharani et al. / Procedia Technology 24 (2016) 1374 – 1387

The images in the first and second row are original brain tumor images and the images in the second and fourth row are the results which represents brain tumor. After applying the original image to the algorithm only tumor region is enhanced and rest of the part is segmented. Table.1. brain area results Patient

Tumor area in pixels A

Tumor Area in sq.mm

B=(A*0.071289)

Brain Area in pixels C

Brain Area in sq.mm

D=(C*0.071289)

Ratio of Areas

G(in sq.mm)

P1

474

33.791

17210

1226.884

0.028

p2

1413

100.731

23782

1695.395

0.059

P3

5930

422.744

26109

1861.285

0.227

P4

4107

292.784

27284

1945.049

0.151

P5

372

26.520

34220

2439.510

0.011

P6

2012

143.433

22197

1582.402

0.091

P7

2143

152.772

35255

2513.294

0.061

P8

677

48.263

33008

2353.107

0.021

P9

868

61.879

30950

2206.395

0.028

P10

647

46.124

17445

1243.637

0.037

P11

721

51.399

30725

2190.355

0.023

P12

2334

166.389

29823

2126.052

0.078

P13

341

24.310

29823

2126.052

0.011

P14

164

11.691

29823

2126.052

0.005

P15

1620

115.488

10513

749.461

0.154

P16

2432

173.375

34280

2443.787

0.071

P17

7264

517.843

26687

1902.490

0.272

P18

647

46.124

17533

1249.910

0.037

P19

2562

182.642

37080

2643.396

0.069

P20

6628

472.503

36645

2612.385

0.181

P21

4107

292.784

27081

1930.577

0.152

P22

5920

422.031

30141

2148.722

0.196

P23

1933

137.802

34898

2487.844

0.055

P24

1163

82.909

19493

1389.636

0.060

P25

377

26.876

15959

1137.701

0.024

P26

1188

84.691

20094

1432.481

0.059

P27

217

15.470

16058

1144.759

0.014

P28

1926

137.303

18787

1339.306

0.103

P29

1926

137.303

37308

2659.650

0.052

P30

1163

82.909

22097

1575.273

0.053

P31

311

22.171

30141

2148.722

0.010

P32

1309

93.317

29180

2080.213

0.045

1382

P33

K. Sudharani et al. / Procedia Technology 24 (2016) 1374 – 1387

2008

143.148

24430

1741.590

0.082

P34

825

58.813

30543

2177.380

0.027

P35

1157

82.481

29315

2089.837

0.039

P36

4089

291.501

24299

1732.251

0.168

P37

774

55.178

21547

1536.064

0.036

P38

682

48.619

21547

1536.064

0.032

P39

729

51.970

21547

1536.064

0.034

P40

1027

73.214

13867

988.565

0.074

P41

2361

168.313

20890

1489.227

0.113

P42

123

8.769

13176

939.304

0.009

P43

997

71.075

14262

1016.724

0.070

P44

257

18.321

27879

1987.466

0.009

P45

494

35.217

14664

1045.382

0.034

P46

669

47.692

18132

1292.612

0.037

P47

374

26.662

11498

819.681

0.033

P48

3405

242.739

15208

1084.163

0.224

P49

1174

83.693

31988

2280.393

0.037

P50

724

51.613

26959

1921.880

0.027

P51

706

50.330

18915

1348.431

0.037

P52

1610

114.775

16936

1207.351

0.095

P53

718

51.186

18628

1327.971

0.039

P54

1019

72.643

24868

1772.815

0.041

Fig.3. Bar graph of tumor images indicating the area of each image applied to the proposed algorithm

1383

K. Sudharani et al. / Procedia Technology 24 (2016) 1374 – 1387

3.1. Mathematical calculations The mathematical calculations of Sensitivity, Specificity, Accuracy and Similarity Index are presented in this section. Before calculating these mentioned parameters one should know few important terms and their definitions which are standards for calculating the above mentioned parameters and many computerized methods can also be presented in [19] but it has complex methodology which is not understandable easily. x

True positives (TP): Brain tumor images are correctly recognized. Which means the people who has brain tumor are correctly identified.

x

False positives (FP): Non-Brain tumor images are incorrectly recognized. This indicates the people who do not have brain tumor are incorrectly identified as they have brain tumor. Simply Healthy people incorrectly identified as sick.

x

True negatives (TN): Non-Brain tumor images are correctly recognized as they do not have brain tumor. Easily Healthy people correctly identified as healthy.

x

False negatives (FN): Brain tumor images are incorrectly recognized. Which represents the people who has brain tumor are incorrectly identified as they do not have brain tumor. Sick people incorrectly identified as healthy. *100%

(9)

*100% (10)

*100% (11) *100% (12) ¾

To calculate above mentioned parameters we need tumor images and no tumor images.

Total tumor images tested=75 Total tumor images tested=45 TP=40, FN=5 Total no tumor images tested= 10 TN= 9, FP=1

PARAMETER Sensitivity Specificity Accuracy Similarity index

Table.2 Results of the above mentioned mathematical calculations VALUE (IN %) 88.9 90 89.2 93.02

1384

K. Sudharani et al. / Procedia Technology 24 (2016) 1374 – 1387

Fig.4 Bar graph of the calculations

4.

Ground truth images

Since in lots of cases the effect of image processing algorithms cannot be evaluated by device or existing objective indicators, ground truth is often generated manually by human experts in the corresponding fields. However, the experts sometimes are not available for the algorithms testing. For instance, in medical image processing we need to segment the ROI , but the doctor is not around to confirm if the ROI is correct. Thus the ground truth image database is presented for the convenience.

K. Sudharani et al. / Procedia Technology 24 (2016) 1374 – 1387

Fig.5 ground truth images and their result. Table.3 ground truth images representing their area Ground Truth Images Area S.No

Images

1

GroungTruth1

Area pixels 1500

in

Area in mm

2

GroungTruth2

1330

351.8953

3

GroungTruth3

3077

814.1218

4

GroungTruth4

2144

567.2659

5

GroungTruth5

61892

16375.5710

6

GroungTruth6

284

75.1415

7

GroungTruth7

1270

336.0204

8

GroungTruth8

370

97.8957

9

GroungTruth9

954

252.4121

10

GroungTruth10

2675

707.7595

11

GroungTruth11

929

245.7976

12

GroungTruth12

251

66.4103

13

GroungTruth13

61182

16187.7171

14

GroungTruth14

3058

809.0948

15

GroungTruth15

3104

821.2656

396.8745

1385

1386

K. Sudharani et al. / Procedia Technology 24 (2016) 1374 – 1387

Fig.6 Bar graph of ground truth images indicating their area.

5.

Conclusion

A new approach for automatic segmentation of tumors from T2 weighted MRI images is presented in this paper. The algorithm also works greatly with the ground truth images and even if the images are with the low intensity regions can be detected with this methodology. After detecting the tumor region, certain standard calculations are carried out, which presented in the paper with the same algorithm, the calculations will help in the process of diagnosing the tumor. Results show that our newly proposed method significantly out-performs. References [1] Bezdek JC, Hall LO, Clarke LP, 1993. Review of MR image segmentation techniques using pattern recognition. Med Phys; 20,4:1033-48. [2] Brunberg JA, Chenevert TL, McKeever PE, Ross DA, Junck LR, Muraszko KM, Dauser R, Pipe JG, Betley AT,1995:In vivo MR determination of water diffusion coefcients and diffusion anisotropy: correlation with structural alteration in gliomas of the cerebral hemispheres. Am J Neuroradiol 16:361-371. [3] W.B. Dou, S. Ruan, Y.P. Chen, D. Bloyet, J.M. Constans, 2007. A framework of fuzzy information fusion for the segmentation of brain tumor tissues on MR images, Image and Vision Computing 25,2: 164–171. [4] Vannier MW, Speidel CM, Rickmans DL,1988. Magnetic resonance imaging multispectral tissue classification. News Physiol Sci;3,1: 48-54. [5] Clark MC, Hall LO, Goldgof DB, 1994. MRI segmentation using fuzzy clustering techniques: integrating knowledge. IEEE Eng Med Biol 13,5:730-742. [6] Ed-Edily Mohd. Azhari, Muhd. Mudzakkir Mohd. Hatta,2014,Brain Tumor Detection And Localization In Magnetic Resonance Imaging. IJITCS: 4,1:1-11. [7] V.J. Nagalkar, S.S Asole,2012; Brain Tumor Detection Using Digital Image Processing Based On Soft Computing. Journal Of Signal And Image Processing:3, 3:102-105. [8] Peck D, Windham J, Emery L, Soltanian-Zadeh H, Hearshen D, Mikkelsen T, 1996. Cerebral tumor volume calculations using planimetric and eigen image analysis. Med Phys;23,12: 2035-2042. [9] Lynn M. Fletcher-Heatha, Lawrence O. Halla, Dmitry B. Goldgofa, F. Reed Murtaghb,2001: Automatic segmentation of non-enhancing brain tumors in magnetic resonance images:Artificial Intelligence in Medicine:21,

K. Sudharani et al. / Procedia Technology 24 (2016) 1374 – 1387

1–3; 43-63. [10] D. Shevad, 2005: Multiple object segmentation and tracking in image sequence. Lamar University -Beaumont. [11]. P.Vasuda, S.Satheesh, 2010 “Improved Fuzzy C-Means Algorithm for MR Brain Image Segmentation”, International Journal on Computer Science and Engineering, 02, 05:1713-1715. [12] Prastawa M, Bullitt E, Ho S, Gerig G, 2004.A brain tumor segmentation framework based on outlier detection”, Med Image Anal, 8, 3;275-283. [13] Yu Sun, Bir Bhanu, Shiv Bhanu, 2009 ;Automatic Symmetry-integrated Brain Injury Detection in MRI Sequences; IEEE conference 79 - 86, DOI: 10.1109/CVPRW.2009.5204052. [14] Salman YL, Assal MA, Badawi AM, Alian SM and MEI-EI Bayome ,2005; Validation techniques for quantitative brain tumors measurements. IEEE conference,7048-7051,DOI: 10.1109/IEMBS.2005.1616129. [15] Kotkar, V.A. ; Gharde, S.S. 2013; Image contrast enhancement by preserving brightnessusing global and local features, Computational Intelligence and Information Technology, IEEE International Conference, 262-271, DOI: 10.1049/cp.2013.2601 [16] Nooshin Nabizadeh, Nigel John, Clinton Wright, 2014. Histogram-based gravitational optimization algorithm on single MR modality for automatic brain lesion detection and segmentation,Expert Systems with Applications; 41, 17; 7820-7836. [17] Manoj k kowar , sourabh Yadav, 2012; Brain Tumor Detction and Segmentation Using Histogram Thresholding;IJEAT, 1,4; 16-20. [18] Somasundaram, K , Gayathri, S.P,2012. Brain segmentation in magnetic resonance images using fast fourier transform, Emerging Trends in Science, Engineering and Technology (INCOSET), IEEE International Conference,164-168, DOI: 10.1109/INCOSET.2012.6513899. [19] El-Sayed A. El-Dahshan, Heba M. Mohsen, Kenneth Revett, Abdel-Badeeh M. Salem 2014:Computer-aided diagnosis of human brain tumor through MRI: A survey and a new algorithm . Expert Systems with Applications; 41,11; 5526-5545. [20] Lynn M. Fletcher-Heath, Lawrence O. Hall,Dmitry B. Goldgof, F. Reed Murtagh, 2001;Automatic segmentation of non-enhancing brain tumors in magnetic resonance images. Artificial Intelligence in Medicine. 21 ,1; 43-63 .

1387