Brain Tumour Segmentation and Measurement Based on Threshold and Support Vector Machine Classifier

 

Ganesan P¹*, B. S. Sathish², V. Elamaran³, R. Murugesan⁴

¹Department of Electronics and Communication Engineering, Vidya Jyothi Institute of Technology,

Hyderabad, India

²Department of Electronics and Communication Engineering, Ramachandra College of Engineering, Eluru, India

³Department of Electronics and Communication Engineering, School of EEE,

SASTRA Deemed University, Thanjavur, India

⁴Department of Electronics and Communication Engineering, Narsimha Reddy Engineering College,  Secunderabad, Telangana, India

 

ABSTRACT:

In medical image processing, the detection and segmentation of brain tumor from the test image is the challenging but most important task. According to WHO, the brain tumor is second most common of cancer death among young people. The proposed approach for the brain tumor segmentation is worked on the principle of threshold and classification using support vector machine. The brain image of the brain is processed in such a way so that the tumor is extracted and displayed the segmented the tumor portion of the image. The gray level co-occurrence matrix and other image quality measures are utilized to calculate the statistical features of the extracted tumor portion of brain. Based on the extracted features, the category of tumor, either benign or malignant, is classified using SVM classifier.

 

 

1. INTRODUCTION:

Brain tumor is the anomalous and abnormal growth of superfluous cells in brain⁵. We know that for all living things the cell is the fundamental unit for its structure. Our body itself produces newer cells to replace the damaged or old ones. The unnecessary, unregulated growth of cells leads to the formation of tumor^1-2.  According to the nature of the tumor, it can be classified as primary (main) or secondary (metastatic). It is very difficult to define the degree of malignancy or aggressiveness of a brain tumor i.e., to sort a brain tumor as “benign” or “malignant” as several issues other than pathological aspects⁴.

 

Metastatic (secondary) brain tumors cancer cells that initiate growing in any part of the body and stretch to the brain. For instance, breast or lung cancer cells are often spread to the brain via the bloodstream. All metastatic brain tumors are malignant (brain cancer). Benign tumor has slow growth rate and less harm. It has distinct borders and rarely spreads¹¹.

 

Even though it is less harm, Surgery is the effective solution for this problem. The malignant tumor growth rate is rapid and uncontrolled one. For this life- threatening and invasive tumor, surgery is the one but effective solution. The assignment of the grade to the tumor is based on the appearance of the tumor cells using some characteristic measures of the tumor cells such as rate of the growth, appearance (similarity to normal cells),  dead tumor cells in the center of the tumor (especially grade four), blood supply and invasive potential. According to WHO, tumors are categorized into four grades based on rate of growth, blood supply, dead cells, uncontrolled growth and healthy cells. The following table illustrates the characteristics of various tumors.

 

First grade tumors are the smallest amount of malignant and tumors grow very slowly. These tumors have a common appearance and only surgery is the effectual treatment for this grade of tumor. The second-grade tumors have slow growth and have a slightly abnormal. There is no clear demarcation between second and third grade tumors. Third grade tumor cells are aggressively replicating the uncharacteristic cells which develop into nearby normal brain tissue.

 

Table 1: Brain Tumor Classification

Characteristics	Grade I	Grade II	Grade III	Grade IV
Malignant level	Smallest amount of malignant	Fair	Moderate	Mostly
Growth rate	Slow	Relatively slow	Active	Abnormal cells which reproduce rapidly
Appearance	Almost looks like normal brain cells	Slightly abnormal. The tumor cells look less like normal cells than first Grade ceils.	Abnormal i.e., tumor cells appearance is very different from normal cells	Most abnormal
Harmness	Less Harm	Possibilities of recurring as higher-grade tumors	Chances to recur as higher-grade tumor	Extremely harmful
Spread	Least preferred malignant	Possibly invade adjacent tissues	Penetrate to adjacent tissues	Maintains rapid growth
Tissue	Tissue is benign	Tissue is malignant	Malignant tissue	Most Malignant
Example	Pilocytic astrocytoma, Gangliocytoma Craniopharyngioma and Ganglioglioma	Astrocytoma, Oligodendroglioma, Ependymoma	Astrocytoma, Oligodendroglioma, Ependymoma	Glioblastoma, Medulloblastoma

 

Fourth grade tumors are considered as most dangerous malignant tumors. These tumor cells appearances is very different from normal cells and reproduce rapidly.  These cells simply develop into neighboring normal brain tissue. They also have areas of dead cells in their center. A metastatic brain tumor is formed by primary cancer cells from any part of the body (for example, lung, colon, kidney and breast) which stretch to the brain. The risk factors for brain tumors can broadly be classified into either environmental or genetic. The environmental factors includes exposed to poisonous substances (air pollution, chemicals etc.), smoking and consuming alcohol, working in chemical or petroleum industries, experiencing viruses and common infections, having a history of head trauma, epilepsy, seizures or convulsions and using common medications like birth control pills, sleeping pills, headache remedies, over-the-counter pain treatments and antihistamines. The segmentation is the process of clustering the complete image into smaller regions ^{3, 5-9}. All regions are homogeneous w.r.t any one characteristics of image such as color texture or intensity ^{10- 15}. There are lot of approaches for clustering. Most of the methods are application oriented ^17-20. The segmentation plays a greater role in the image analysis because the outcome of the segmentation is the basis for image analysis ^24-28. There are lot of methods and techniques to detect and segment the brain tumor from the input test image. Based on the techniques applied, they divided into edge based, region based, threshold based and cluster-based techniques. As far as edge-based method is concerned, the harsh variations in the intensity values is the major concern for the segmentation of the tumor.  In threshold-based techniques, the success of the brain tumor extraction is based on the selection of the proper threshold value. In region-based techniques, the brain tumor image is clustered into number of regions having dissimilar uniqueness. When it comes to clustering based brain tumor detection and segmentation, brain tumor image is divided into number of segments or clusters based on the image characteristics such as texture, color or intensity. The classification based on the allocation of same group of pixels into the same cluster.

 

2. METHODOLOGY:

The proposed approach for the segmentation and detection of brain tumor is explained as follows.

·       Test image is obtained from image database which is collection of brain tumor images.

·       Pre-processing of the test image. If the image is in RGB color space, the image should be transformed to its gray scale version. The image also resized to 250 * 250 to the efficient analysis of the tumor images.

·       The image enhancement operation is performed on the pre-processed image to remove the noisy pixels.

·       The intensity image is converted into binary image using global threshold method, Otsu²¹ (threshold value is 0.5 as default).

·       Removes all connected components (objects) from a binary image that have less than 100 pixels using the operation called area opening²⁴ (bwareaopen command of MATLAB).

·       Principal component analysis (PCA) exploits an orthogonal transformation to convert a set of correlated observations into a set of linearly uncorrelated variables (values) called principal components

·       Create gray-level co-occurrence matrix (GLCM) from image.

·       Compute the properties of gray-level co-occurrence matrix

·       Train SVM classifier with training data.

 

In MATLAB, The function svmtrain is utilized to determine the support vectors (s_i), weights (α_i), and bias (b). They are applied to categorize vectors (x) based on the following equation:

 

here k=a kernel function. If c ≥ 0, then x is categorized as an element of the group one, if not it is categorized as an element of another group.

Categorize whether the tumor is malignant or benign using support vector machine (SVM).

 

Fig. 1 Computation of gray-level co-occurrence matrix form image matrix (Image Courtesy: Mathworks)

 

 

Fig. 2Work flow of the proposed approach

 

(a) input image                                      (b) Filtered Image                                    (c) Enhanced Image

 

     

Fig 3: The outcome of the proposed approach for the brain tumor detection and segmentation

 

3. EXPERIMENTAL ANALYSIS:

The investigational outcome of the proposed method of brain tumor detection and classification from the test image is depicted in fig 3. A database, collection of brain tumor images, is created. Test image is obtained from image database. If the image is in RGB color space, the image should be transformed to its gray scale version. The image also resized to 250* 250 to the efficient analysis of the tumor images. The image enhancement operation is performed on the pre-processed image to remove the noisy pixels. This is shown in fig 3(c). The original image is divided into our quadrants as shown in fig 3(d). This operation is for the clear demarcation of the tumor from other portion of the image. The detected tumor portion is depicted in fig 3(e). The tumor portion is thresholded and extracted portion of the tumor is displayed in fig 3(g). The measurement on the extracted tumor portion is illustrated in table 2. Principal component analysis (PCA) exploits an orthogonal transformation to convert a set of correlated observations into a set of linearly uncorrelated variables (values) called principal components. It is necessary to construct GLCM (gray-level co-occurrence matrix) from image to compute the features (quality measures) such as contrast, homogeneity, correlation and energy. The type of the tumor (malignant or benign) is classified using support vector machine (SVM).

 

Table 2: Brain Tumor Measurement

4. CONCLUSION:

The segmentation o the brain tumor based on threshold and classification using support vector machine is explained. The segmentation process is performed by global threshold, method and image area opening is applied to remove the small components form the tumor portion. The comprehensive measurement of tumor is clearly depicted in the result analysis. SVM classifier is finally classified the tumor, either benign or malignant, based on the extracted features. The proficiency of the proposed method for the segmentation of the brain tumor from the test image is clearly demonstrated based on the outcome of the experimental result.

 

5. REFERENCES:

1.      S. Pereira, A. Pinto, V. Alves and C. A. Silva, "Brain Tumor Segmentation Using Convolutional Neural Networks in MRI Images," in IEEE Transactions on Medical Imaging, vol. 35, no. 5, pp. 1240-1251, May 2016.

2.      S. Bauer, "A survey of MRI-based medical image analysis for brain tumor studies", Phys. Med. Biol., vol. 58, no. 13, pp. 97-129, 2013.

3.      Kalist V, Ganesan P, Sathish BS, and Jenitha JMM. Possiblistic-Fuzzy C-Means Clustering Approach for the Segmentation of Satellite Images in HSL Color Space.  Procedia Computer Science. 57; 2015; 49-56.

4.      J.J. Corso, E. Sharon, S. Dube, S. El-Saden, U. Sinha, A. Yuille, "Efficient multilevel brain tumor segmentation with integrated bayesian model classification", IEEE Transactions on Medical Imaging, vol. 27, no. 5, pp. 629-640, 2008.

5.      Ganesan, P. and Rajini, V., “Segmentation and Comparison of Water Resources in Satellite Images using Fuzzy based Approach”, Advances in Intelligent Systems and Computing (ISSN 2194-5357), Advances in Soft Computing, Springer Verlag, Vol. 308, No. 1. pp 685-692, 2015.

6.      Ganesan, P. and Rajini, V., “Unsupervised Segmentation of Satellite Images based on Neural Network and Genetic Algorithm”, Advances in Intelligent Systems and Computing (ISSN 2194-5357), Advances in Soft Computing, Springer Verlag, Vol. 309, No. 2, pp 319-326, 2015

7.      R. A. Heckemann, J. V. Hajnal, P. Aljabar, D. Rueckert, A. Hammers, "Automatic anatomical brain MRI segmentation combining label propagation and decision fusion", Neuro Image, vol. 33, pp. 115-126, 2006.

8.      Shaik KB, Ganesan P, Kalist V, and Sathish BS. Comparative Study of Skin Color Detection and Segmentation in HSV and YCbCr Color Space.  Procedia Computer Science. 57; 2015; 41-48.

9.      P. Ganesan, B. S. Sathish and G. Sajiv, "A comparative approach of identification and segmentation of forest fire region in high resolution satellite images," 2016 World Conference on Futuristic Trends in Research and Innovation for Social Welfare (Startup Conclave), Coimbatore, 2016, pp. 1-6. doi: 10.1109/STARTUP.2016.7583959

10.   Ganesan, B. S. Sathish, K. B. Shaik and V. Kalist, "Neural network-based SOM for multispectral image segmentation in RGB and HSV color space," 2015 International Conference on Circuits, Power and Computing Technologies [ICCPCT-2015], Nagercoil, 2015, pp.1-6.doi: 10.1109/ICCPCT.2015.7159345

11.   M. Huang, W. Yang, Y. Wu, J. Jiang, W. Chen and Q. Feng, "Brain Tumor Segmentation Based on Local Independent Projection-Based Classification," in IEEE Transactions on Biomedical Engineering, vol. 61, no. 10, pp. 2633-2645, Oct. 2014.

12.   Ganesan P and B. S. Sathish. Automatic Detection of Optic Disc and Blood Vessel in Retinal Images using Morphological Operations and Ipachi Model. Research J. Pharm. and Tech. 10(8): August 2017; 2602-2607. pp.35-41.

13.   Ganesan, P and Palanivel, K and Sathish, BS and Kalist, V and Shaik, Khamar Basha, “Performance of fuzzy based clustering algorithms for the segmentation of satellite images-A comparative study”, IEEE Seventh National Conference on Computing, Communication and Information Systems (NCCCIS), 2015, pp 23 – 27.

14.   S. Bauer, L. P. Nolte, M. Reyes, "Fully automatic segmentation of brain tumor images using support vector machine classification in combination with hierarchical conditional random field regularization", Proc. Med. Image Comput. Comput. Assist. Interv., pp. 354-361, 2011.

15.   Ganesan Pand Shaik KB. HSV color space-based segmentation of region of interest in satellite images.  2014 International Conference on Control, Instrumentation, Communication and Computational Technologies (ICCICCT). 2014; 101-105.doi: 10.1109/ICCICCT.2014.6992938

16.   Wulandari, R. Sigit and M. M. Bachtiar, "Brain Tumor Segmentation to Calculate Percentage Tumor Using MRI," 2018 International Electronics Symposium on Knowledge Creation and Intelligent Computing (IES-KCIC), Bali, Indonesia, 2018, pp. 292-296.

17.   Ganesan P and Shaik KB. HSV color space-based segmentation of region of interest in satellite images. 2014 International Conference on Control, Instrumentation, Communication and Computational Technologies (ICCICCT). 2014; 101-105.doi: 10.1109/ICCICCT.2014.6992938

18.   Huang Meiyan, Wei Yang, Wu Yao, Jiang Jun, Chen Wufan, Qianjin Feng, "Brain Tumor Segmentation Based on Local Independent Projection-based Classification", IEEE Transactions on Biomedical Engineering, 2013.

19.   D. Bhattacharyya, T. H. Kim, "Brain tumor detection using MRI image analysis", Commun. Comput. Inform. Sci., vol. 151, pp. 307-314, 2011.

20.   Ganesan P, V Rajini, BS Sathish, V Kalist, SK Khamar Basha, Satellite Image Segmentation Based on Ycbcr Color Space. Indian Journal of Science and Technology. Vol 8 Issue 1, (2015), pp 35-41

21.   C. L. Biji, D. Selvathi, A. Panicker, "Tumor detection in brain magnetic resonance images using modified thresholding techniques", Commun. Comput. Inform. Sic., vol. 4, pp. 300-308, 2011.

22.   Ganesan P, M.Ganesh , L.M. I. Leo Joseph and V. Kalist, “ Central Retinal Vein Occlusion: An Approach for the Detection and Extraction of Retinal Blood Vessels”, J. Pharm. Sci. & Res. Vol. 10(1), 2018, 192-195.

23.   T. M. Hsieh, Y. M. Liu, C. C. Liao, F. Xiao, I. J. Chiang, J. M. Wong, "Automatic segmentation of meningioma from non-contrasted brain MRI integrating fuzzy clustering and region growing", BMC Med. Informat. Decision Making, vol. 11, pp. 54, 2011.

24.   https://www.mathworks.com/help/matlab/

25.   Ganesan P, “Detection and Segmentation of Retinal Blood Vessel in Digital RGB and CIELUV color space Fundus Images”, Research J. Pharm. and Tech. 11(6): 2018, 2326-2330.

26.   Sajiv G and Ganesan P. Comparative Study of Possiblistic Fuzzy C-Means Clustering based Image Segmentation in RGB and CIELuv Color Space. International Journal of Pharmacy & Technology. 8(1); 2016; 10899-10909.

27.   Ganesan P and Sajiv G. Unsupervised Clustering of Satellite Images in CIELab Color Space using Spatial Information Incorporated FCM Clustering Method. International Journal of Applied Engineering Research. 10(20); 2015.

28.   Sathish BS, Ganesan P and Khamar Basha. Shaik. Color Image Segmentation based on Genetic Algorithm and Histogram Threshold. International Journal of Applied Engineering Research. 10(6); 2015; 123-127.

 

Accepted on 14.12.2019           © RJPT All right reserved

Research J. Pharm. and Tech 2020; 13(6):2573-2577.