Classification of Brain Cancer Using Artificial Neural Network

Dipali M. Joshi, Dr.N. K. Rana Department of Instrumentation.

Rajiv Gandhi Institute of Technology Mumbai, India e.mail:dipaishaj@gmail.com, ranank@rediffmail.com
Abstract - A Brain Cancer Detection and Classification System has been designed and developed. The system uses computer based procedures to detect tumor blocks or lesions and classify the type of tumor using Artificial Neural Network in MRI images of different patients with Astrocytoma type of brain tumors. The image processing techniques such as histogram equalization, image segmentation, image enhancement, morphological operations and feature extraction have been developed for detection of the brain tumor in the MRI images of the cancer affected patients. The extraction of texture features in the detected tumor has been achieved by using Gray Level Co-occurrence Matrix (GLCM). These features are compared with the stored features in the Knowledge Base. Finally a Neuro Fuzzy Classifier has been developed to recognize different types of brain cancers. The whole system has been tested in two phases firstly Learning/Training Phase and secondly Recognition/Testing Phase. The known MRI images of affected brain cancer patients obtained from Radiology Department of Tata Memorial Hospital (TMH) were used to train the system. The unknown samples of brain cancer affected MRI images are also obtained from TMH and were used to test the system. The system was found efficient in classification of these samples and responds any abnormality. Keywords: Brain cancer; MRI; Co-occurrence Matrix; ANN

V. M. Misra.
Department of Instrumentation. Vivekanand Education Societys Institute of Technology, Mumbai, India e.mail:vmisra2004@yahoo.com, of cancer affected patients. The system also finds sufficient usage under Cancer Detection in the area of medical sciences such as Computer Aided Diagnosis and Mammography etc. Brain cancer is a complex disease, classified into 120 different types. So called non malignant (Benign) brain tumors can be just as life-threatening as malignant tumors, as they squeeze out normal brain tissue and disrupt function. The glioma family of tumors comprises 44.4 % of all brain tumors. Glioblastoma type of Astrocytoma is the most common glioma which comprises 51.9 %, followed by other types of astrocytoma at 21.6 % of all brain tumors [1, 2]. Brain tumors are the leading cause cancer death in children under the age of 20. They are the second leading cause of cancer death among 20-29 year old males. Metastatic brain tumors result from cancer that spreads from other parts of the body into the brain. About 10-15 % of people with cancer will eventually develop metastatic brain tumors [1]. There are many types of Brain Cancers but among various types of Brain Cancers the most prevalent and common is Astrocytoma. The method of detection and work carried out during this study is explained in brief in next section. The image classification and feature extraction techniques are discussed in section III. The results obtained during experimentation and conclusions drawn are elaborated in section IV and V respectively. II. METHOD

I.

INTRODUCTION

Curing cancer has been a major goal of medical researchers for decades, but development of new treatments takes time and money. . Science may yet find the root causes of all cancers and develop safer methods for shutting them down brain tumors are benign and can be before they have a chance to grow or spread. Approximately 40 percent of all primary successfully treated with surgery and, in some cases, radiation [1].The number of malignant brain tumors appears to be increasing but for no clear reason. Magnetic Resonance Imaging (MRI) has become a widely used method of high quality medical imaging, especially in brain imaging where MRIs soft tissue contrast and noninvasiveness is a clear advantage. MRI provides an unparalleled view inside the human body. The level of detail we can see is extraordinary compared with any other imaging modality. Reliable and fast detection and classification of brain cancer is of major technical and economical importance for the doctors. Common practices based on specialized technicians are slow, have low responsibility and possess a degree of subjectivity which is hard to quantify. The designed system is an efficient system for Detection and Classification of Brain Cancer from a given MRI image

The work carried out involves processing of MRI images that are affected by brain cancer for detection and Classification on different types of brain tumors. The image processing techniques like histogram equalization, image segmentation, image enhancement and then extracting the features using Gray Level Co-occurrence Matrix are used for Detection of tumor. Extracted feature are stored in the knowledge base. A suitable Neuro Fuzzy Classifier is developed to recognize the different types of brain cancers. Images used are MRI images obtained from Tata Memorial hospital. The system is designed to be user friendly by creating Graphical User Interface (GUI).

978-1-4244-7406-6/10/$26.00 c 2010 IEEE

112

Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on June 23,2010 at 08:16:42 UTC from IEEE Xplore. Restrictions apply.

Standard MRI Image

New MRI Image

All the slices processed by the system have been automatically classified as abnormal. They are known to contain Astrocytoma tumor based on radiologist pathology report. B. Stage one: Tumor Segmentation. The first step in the system presented here is to isolate the tumor region from the rest of the image. Various image processing techniques are used to separate the tumor region. Image preprocessing consists mainly of Histogram Equalization. The main problem in the process of detection of edge of tumor is that the tumor appears very dark on the image which is very confusing. To overcome this problem, Histogram Equalization was performed. Segmentation subdivides an image into its constituent parts or objects. The level to which this subdivision is carried depends on the problem being solved, that is, the segmentation should stop when the edge of the tumor is able to be detected, and the main interest is to isolate the tumor from its background [3]. Thresholding has been used for segmentation as it is most suitable for the present application in order to obtain a binarized image with gray level 1 representing the tumor and gray level 0 representing the background. In simple implementations, the segmentation is determined by a single parameter known as the Intensity Threshold. In a single pass, each pixel in the image is compared with this threshold. If the pixel's intensity is higher than the Threshold, the pixel is set to white, in the output. If it is less than the Threshold, it is set to black. The process is described by following equation [4].

Image Preprocessing Image Segmentation Image Enhancement Detection of Tumor Texture Feature Extraction Using Gray Level Co-occurrence Matrix
Features of known samples

Knowledge Base Fuzzification

Features of unknown samples

ANN Based Classifier Result of Classification

Figure 1. Block Diagram of the System

The designed and developed system works in two phases namely Learning/Training Phase and Recognition/Testing Phase. In Learning/Training Phase the ANN is trained for recognition of different Astrocytoma types of brain cancer. The known MRI images are first processed through various image processing steps such as Histogram Equalization, Thresholding, and Sharpening Filter etc. and then textural features are extracted using Gray Level Co-occurrence Matrix. The features extracted are used in the Knowledge Base which helps in successful classification of unknown Images. These features are normalized in the range -1 to 1 and given as an input to Artificial Neural Network Based Classifier. The unknown MRI images affected by cancer of type Astrocytoma are used for testing in Recognition/Testing Phase. III. IMAGE CLASSIFICATION STAGES

The fundamental enhancement needed is to increase the contrast between the whole brain and the tumor. Contrast between the brain and the tumor region may be present but below the threshold of human perception. Thus, to enhance the contrast between the normal brain and tumor region, a sharpening filter is applied to the digitized MRI resulting in noticeable enhancement in image contrast. The dilation operator is used for filling the broken gaps at the edges and to have continuities at the boundaries. Onto the dilated image a filling operator is applied to fill the close contours. After filling operation on an image, centroids are calculated to localize the regions. The final extracted region is then logically operated for extraction of Massive region in given MRI image. C. Stage two: Feature Extraction The work involves extraction of the important features for image recognition. The features extracted give the property of the texture, and are stored in knowledge base

A. Stage zero: Pathological Detection.

2010 2nd International Conference on Electronic Computer Technology (ICECT 2010)

113

Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on June 23,2010 at 08:16:42 UTC from IEEE Xplore. Restrictions apply.

The extracted features are compared with the features of Unknown sample Image for classification. Texture features or more precisely, Gray Level Co- occurrence Matrix (GLCM) features are used to distinguish between normal and abnormal brain tumors. Five co-occurrence matrices are constructed in four spatial orientations horizontal, right diagonal, vertical and left diagonal (0, 45, 90, and 135) [5, 6]. A fifth matrix is constructed as the mean of the preceding four matrices. The extracted features are shown in table 1.
TABLE 1. ASM Contrast EXTRACTED FEATURES IDM Dissim ilarity I/P 1 I/P 2

result, knowledge contained in feature space is extracted and utilized. E. Stage four: Nero-Fuzzy Classifier. A Neuro-Fuzzy Classifier is used to detect candidate circumscribed tumor. ANN'S are networks of interconnected computational units, usually called nodes. The input of a specific node is the weighted sum of the output of all the nodes to which it is connected. The output value of a node is, in general, a non-linear function (referred to as the activation function) of its input value. The multiplicative weighing factor between the input of node j and the output of node i is called the weight wji. An Artificial Neural Network is an adaptive, most often nonlinear system that learns to perform a function (an input/output map) from data. Adaptive means that the system parameters are changed during operation, normally called the Learning/Training phase. After the training phase the Artificial Neural Network parameters are fixed and the system is deployed to solve the problem at hand (The Recognition/Testing phase). Back-propagation ANN's used in this study consist of one input layer, one or two hidden layers, and one output layer [8, 9]. With back-propagation, the input data (Extracted Features) is repeatedly presented to the Artificial Neural Network, with each presentation the output of the neural network is compared to the desired output (Grade of Tumor) and an error is computed. This error is then fed back (back-propagated) to the Artificial Neural Network and used to adjust the weights such that the error decreases with each iteration and the neural model gets closer and closer to producing the desired output. This process is known as Training. The Training of these networks consists in finding a mapping between a set of input values and a set of output values. This mapping is accomplished by adjusting the value of the weights wji; using a learning algorithm, the most popular of which is the generalized delta rule. After the weights are adjusted on the training set, their value is fixed and the ANN's are used to classify unknown input images [8]. The generalized delta rule involves minimizing an error term defined as [9, 10]

Entropy

               

               

               

D . Stage three: Knowledge Base Knowledge is any chunk of information that effectively discriminates one class type from another. In this case, tumor will have certain properties that other brain tissues will not and vice-versa. In the domain of MRI volumes, there are two primary sources of knowledge available. The first is pixel intensity in feature space, which describes tissue characteristics within the imaging system. The second is image/anatomical space and includes expected shapes and placements of certain tissues within the image, such as the fact that CSF lies within the ventricle. The nature of tumors limits the use of anatomical knowledge, since they can have any shape and occupy any area within the brain [3, 7]. As a

In this equation, the index p corresponds to one input vector, and the vectors tp and op are the target and observed output vectors corresponding to the input vector p, respectively. IV. RESULTS

The developed software efficiently classifies the input MRI image of Brain Cancer affected patients into a Grade of Astrocytoma type of Tumor. The MRI images of patients affected by Brain Cancer are used during Recognition/Testing phase. For the input image used for Testing, the system shows the Tumor Region Extracted from the outer skull of brain. The features extracted from this

114

2010 2nd International Conference on Electronic Computer Technology (ICECT 2010)

Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on June 23,2010 at 08:16:42 UTC from IEEE Xplore. Restrictions apply.

region are compared with stored features in Knowledge base. The developed system then classifies the image into a Grade of the tumor for Astrocytoma type of Brain Cancer.
TABLE 2. Sr.No 1 2 3 4 5 6 7 8 Test Image Image 1 Image 2 Image 3 Image 4 Image 5 Image 6 Image 7 Image 8 SUMMARY OF RESULTS Classified grade of image Grade I Grade II Grade III Grade IV Grade III Grade IV Grade III Grade IV Figure 3. Result of Classification

Original grade of Image Grade I Grade II Grade III Grade IV Grade I Grade II Grade III Grade IV

Table 2 gives the details about the original Grade of Astrocytoma type of tumor, the image is affected by and result of Classification shown by the developed system for that image. The result in figure 2 shows the Tumor Region which is extracted from the outer skull. The figure 3 shows that the given image is affected by the tumor of Grade III type.

The above results show that the system works efficiently for Detection and Classification of Brain cancer. Achieved results of Classification of unknown Brain Cancer Images have been shown and discussed with cancer specialist from TMH. The test results are in agreement with the opinion of the Doctors and provide a confirmation test for cancer detection. V. CONCLUSION

Figure 2.

Extracted region without outer skull

Presently available systems can only detect location and size of tumor but do not provide any knowledge about the type of tumor. Many cancer forms can only be diagnosed after a sample of suspicious tissue has been removed and tested (biopiesd). Pathologists view pathologic tissues, typically with brig field microscopes, to determine the degree of normalcy versus disease. This process is time consuming, and fatiguing. The induced exhaustion created by this process may contribute to diagnostic errors. The system described in this project Classifies and Identifies pathological tissues in an automated fashion. The designed and implemented system provides precision detection and real time tracking by classifying the unknown sample Image into appropriate Astrocytoma type of Cancer, thus do not involve any pathological testing. Brain Cancer Detection and Classification System are implemented using Artificial Neural Network. The design based on Image processing Techniques, Artificial Neural Network and Graphical User Interface was successfully completed and used in the system to Detect and Classify the Tumor. The designed Brain Cancer Detection and Classification System use conceptually simple Classification method using the Neuro Fuzzy logic. Texture features are used in the Training of the Artificial Neural Network. Cooccurrence matrices at different directions are calculated and Grey Level Co-occurrence Matrix (GLCM) features are extracted from the matrices. The above procedure effectively classifies the tumor types in brain images taken under

2010 2nd International Conference on Electronic Computer Technology (ICECT 2010)

115

Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on June 23,2010 at 08:16:42 UTC from IEEE Xplore. Restrictions apply.

different clinical circumstances and technical conditions, which were able to show high deviations that clearly indicated as abnormalities in area with brain disease. The images in this study are acquired from MRI samples obtained from Radiology Department of Tata Memorial Hospital. This system provides precision Detection and Classification of Astrocytoma type of cancer. The system has been tested only with the above sample Images. The system can be designed to classify other types of cancers as well with few modifications. The scope of the system can further be improved by using other types (e.g. PET, MRS, CTS) of Images. It is essential to use large number of patients data which will improve the accuracy of the system. More features that could be added to the system include metabolic and genetic data as well as anatomical attributes of the brain. REFERENCES
[1] Phooi-Yee LAU and Shinji OZAWA, A Simple Method for Detecting Tumor in T2-Weighted MRI Brain Images: An Image-Based Analysis, Department of Information and Computer Science, Keio University, Yokohama-shi, pp-223-8522 Japan

[2] Cline HE, Lorensen E, Kikinis R, Jolesz, Three-dimensional segmentation of MR images of the head using probability and connectivity F. J Computer Assist Tomography 1990; pp 14:10371045. [3] .Mathew C. Clark, Lawrence O. Hall, Automatic tumor segmentation using Knowledge based techniques, IEEE transaction on medical imaging, Vol 17, No. 2, April 1998. [4] X. Descombes, F. Kruggel, G. Wollny, and H.J. Gertz, An objectbased approach for detecting small brain lesions: Application to Virchow-robin spaces, IEEE Trans Med. Imaging, vol.23, no.2, pp.246-255, 2004. [5] Sonka, M. Hlavac, V. Boyle, R. Image processing, Analysis, and Machine Vision, (2004). II Edition, Vikas Publishing House, New Delhi pp.821 [6] Gonzalez, R.C. Richard, E.W; Digital Image Processing, (2004), II Indian Edition, Pearson Education, New Delhi, India. pp.793 [7] Mehmed Ozkan, Benoit M. Dawant, Maciunas and Robert J. Member, IEEE, Neural Network Based Segmentation of Multi-Modal Medical Images: A Comparative and Prospective Study, IEEE transactions on medical imaging, Vol. 12, No. 3, September 1953 [8] Jaceck Zurada, Introduction to Artficial neural systems, Jaico Publishing pages 790. [9] Simon Haykin, Neural Network Design. (2004). I Edition, Vikas Publishing House, New Delhi, India pp. 938 [10] S. N. Sivanandam, S. Sumathi, S. N. Deepa, Introduction to Neural Networks using Matlab 6.0. Tata McGraw Hill Company Ltd, New Delhi, June 2005.

116

2010 2nd International Conference on Electronic Computer Technology (ICECT 2010)

Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on June 23,2010 at 08:16:42 UTC from IEEE Xplore. Restrictions apply.