South Asian Journal of Engineering and Technology Vol.3, No.9 (07) 7 Detection of malignant and benign Tumors by ANN Classification Method K. Gandhimathi, Abirami.K, Nandhini.B Idhaya Engineering College for Women, Chinnasalem, Villupuram District, Tamilnadu. Received: 08/0/07, Revised: 09/0/07 and Accepted: 03/04/07 Abstract Cancer is a very arduous disease to cure. Therefore, observing a tumor at a premature stage can be uncomplicated for treatment and then liberate some people from death. Normally, biomedical imaging supports the doctors for capturing of images for both diagnostic and treatment purposes. This paper suggests the segmentation process to brain MRI image for edge detection of tumor structure by Otsu method. The Otsu permits us to notice a threshold to sector the tumor part and form the binary image of the MRI image. After the threshold detection, the binary image will be labeled with white tumor region for accurately determining the vision of the region in the image. so, the tumor image edge will be established using the unsharp masks, particularly, the edge image will be handled to combine with the original for replacing the segmented image for evaluate of image structure. Keywords MRI brain tumor image set, segmentation, edge detection, threshold. INTRODUCTION There are different types of tumors such as brain, chest and spinal cord, etc. which can appear in human body. According to statistic data, in 07,,688,780 new cancer cases and 600,90 cancer deaths are projected to occur in the United States. Compared to other cancers, brain and other nervous system cancer are relatively rare. However, it is still a challenge for us to detect early for tumor based on medical images for treatment. Different kind of diagnostic imaging techniques such as Single-photon emission computed tomography (SPECT), Positron Emission Tomography (PET), Computed Tomography (CT), Magnetic Resonance Imaging (MRI) [-], in which it may be Benign(non-cancer) or malignant (cancer). In practice, a brain tumor which is early detected is called the primary brain tumor. During its development, the primary tumor may spread to other parts of the brain and is often found problems at somewhere in the body. Moreover, the development of tumor may be lasted and make it to be a metastatic brain tumor. MRI brain images were enhanced using median method for filter of noise before segmentation of the image.therefore,the Otsu method was applied for determination of gray level threshold to the enhanced MRI image based on histogram map calculated. The Otsu allows us to detect a suitable threshold to segment the tumor part of the image [5-6]. In this paper, a set of brain images will be processed to detect the view of the region in the image. II.METHODOLOGY Segmentation of images into meaningful regions has been an important issue in diagnosis of medical image. The D images shown in Fig. will be pre-processed to enhance detection of tumor in MRI brain images. A. Image Transformation In digital image processing, conversion of a color image to grayscale one is to more easily calculate for detection of regions of the gray image. According to National Television Committee (NTSC) standard, RGB pixel values in the color image are converted into gray values using the weighted sum of the R, G, and B components as follows: 7
Y = 0.99R + 0.587G + 0.4B () in which RGB are three basic color components of an color image and Y describes gray level value. B. Sharpened Image Medical images captured using techniques such as CT, MRI or PET often have noise or are burring. For accurately segmenting the medical images, they need to be sharpened, particularly edges around regions of interests are sharpened using an unsharp mask. It means that an image is processed by subtracting a blurred version of the image as described in the following steps:. Pass the original image through Low Pass filter to blur the image.. Subtract the Low Pass filtered image from the original. It will produce the mask. F(x,y) = I(x,y) I (x,y) ----------------- (3) Where, I(x,y) is the original image I (x,y) is the blurred image F(x,y) is the mask 3. Add the mask to the original: G(x,y) = I(x,y) + k*f(x,y) -------------------- (4) here k is a weight. When k = it is unsharp masking; k > it is highboost filtering; when k < it de-emphasize the contribution of a mask. C. Image Segmentation Using Otsu Thresholding Method For detection of objects in an image for analysis and estimation is necessary. An Otsu segmentation method will be employed to find an optimal threshold for detection of tumors in a brain image. This method allows choosing the optimal gray threshold based on the gray histogram map of the image. In particular, after the threshold is chosen using the Otsu method, the obtain image will be the black and white image In Otsu method, one exhaustively searches for the Threshold(T) that minimizes the intra-class variance. It means that this threshold is a weighted sum of variances of the two classes and its equation is expressed as follows: The weighted within-class variance is: q q w Where the class probabilities are estimated as q t ) i P( i q I P( it And the class means are given by: t ip( ( t) ( t) q I ip( q i Finally individual class variances are: t [ i ( t)] i q P( it 8
I [ i ( t)] it q P( After some algebra, we can express the total variance as: q [ q ][ ] w D. Labeled Image In one image, the characteristics of the image regions are different. Therefore, converting a labeled image matrix into the RGB image is performed by assigning each color corresponding to each labeled object. With labeling objects in the segmentation image, each object appears in a different color, so the objects of the segmentation are easier to distinguish than that of the original image. Moreover, labeling for objects of the image enables to easily extract features as well as to detect regions of the view in image. In this paper, the labeling method is applied to determine tumors in MRI brain image. E. Morphological Operation Morphological Operation is the basic operator based on the shape and structure of the image. The Morphological algorithm applies a structural factor for both input image and output image with the same dimensions. Two main morphology operators are dilation and erosion, where the erosion makes the considered objects in the thinner image, while the dilation produces the considered objects in the thicker image. In this Proposed work, the dilation is chosen and its operation chooses the highest value of all neighboring pixels determined by the structuring element. The operator is computed using the following equation: fbx, y max f x x, y y x, y Db F. Edge Detection Using Laplacian Mask For edge detection of image, a Laplacian algorithm is utilized to produce the image with edges of objects. An image f(x,y) with pixel intensity values I(x,y) will be calculated using the following equation: x x, y f x y y f, f x, y III. RESULTS AND DISCUSSION In this research, two brain images with benign and malignant were processed to produce two segmented images for estimation. The images after processing showed regions and textures based on gray levels and the region pixels for diagnosis of tumor type. Simulation results were obtained following the processing steps as described below. A. Enhanced images Two original images of brain tumors captured on a person by the MRI technology as shown in Fig.. The image shows different tumors. The problem is that they need to segment for estimation of size as well as type of two different tumors. Fig..Two original MRI images, in which the left image is the malignant (cancer) brain tumor; the right one is the benign (noncancer )brain tumor. 9
For sharpening images before segmentation, the unsharp mask algorithm was applied to enhance edges of objects in two MRI images with benign and malignant as shown in Fig.. These enhanced images edges more sharpening than compared to that of the original images. Fig.. Sharpened MRI brain images using the unsharp mask The enhanced images was segmented using the Otsu algorithm with the threshold of 0.6 in the range [0 ]. It means that the segmented image is a binary image with the black and white levels assigned [0 ]. Therefore, the threshold value is chosen in the range of [0 ] to detect tumor objects as shown in Fig. 3. From the segmented images, the labeling algorithm was employed for different regions corresponding to the total pixel number of each region. Thus, each region was assigned each different color as shown in Fig. 4. Each color region allows to more easily separating the typical region from the image, particularly the tumors can be separated for diagnosis. Histograms of two images were determined, as shown in Fig. 5, for estimation of the total pixel number of objects to accurately increase in diagnosis of tumors. Fig. 3. Thresholding segmentation of two enhanced MRI brain images using Otsu method with the thresholding of 0.65 with color tumors A morphology operator with dilation and erosion eliminate small objects in the segmented image for exact tumors as shown in Fig. 5. In order to detect edges of these images, a Laplacian wasutilized as shown in Fig. 6. Fig. 4. Two labeled images was applied to determination of algorithm Fig 5.Desired objects of two images were separated Fig. 6. Edge tumor objects in two MRI brain images For exact estimation of brain tumors in medical images, the combination between the original 0
image and segmented image was used. This combination allows us the processed tumor images with the tumor objects which we need to analyze for diagnosis as shown in Fig. 7. Segmentation of medical images allows us to consider regions of the view. Moreover, it is very useful to know components of an object in the image for consideration of image texture for diagnosis of tumor type. As known, tumor in human body is type of solid object, so one needs to consider its feature. In the process of its development, tumor size gradually increases and it can move into the surrounding tissue. Therefore, its edges or surface of the tumor will protrude like convex. This characteristic can be the basic to distinguish tumor type in medical tumor images. Fig. 7. Combination of edge detection and original images A K-mean clustering method was applied to segment MRI brain image for determination of tumor. In this paper, authors divide pixels in the image into clusters and calculate distance from each pixel to each cluster center for finding desires region. Therefore, a morphological filter was employed to enhance the detected tumor. In our research, two MRI brain images with benign and malignant tumors were pre-processed to sharpen the images before image segmentation. For tumor detection, the Otsu approach was applied to find the desired threshold value. Finally, labeled and morphological methods were utilized to separate the desired tumors from the MRI brain images. The result is that with the threshold value of 0.65, the detection of malignant is 90 % per 0 images, while that of benign is 80 % per 0images. IV CONCLUSION In this research, the segmentation of medical image plays an important role in diagnosis and treatment. The article showed development of the segmentation method for detection of MRI brain tumor images. In particular, the MRI brain images were preprocessed to obtain the enhanced image. Therefore, the Otsu segmentation method was employed to detect tumor of the image, in which the tumor can be benign or malignant. This proposed method can support doctors in diagnosis and treatment more exactly and early. As results, the proposed approach applied in this paper illustrated the effectiveness. Another result is that other threshold values are smaller/bigger than 0.65; the percent probability of accurate tumor recognition is much smaller compared to that of the 0.65 threshold. In addition, this proposed segmentation method could be the significant step for development of recognition of benign and malignant tumors of medical images. REFERENCES [] A.H. Foruzan, R. A. Zoroofi, M. Hori, Y. Sato, A knowledge-based technique for liver segmentation in CT data, PubMed journal, Vol. 33, No. 8, pp. 567-87, 009. [] J. E. Mackewn, G. Charles-Edwards and J.J. Totman, P. Halsted, E.J. Somer, S.F. Keevil, P.K. Marsden A fiducial marker based technique for alignment of simultaneously acquired PET and MRI images, proceedings of IEEE Nuclear Science Symposium Conference Record, pp. 3307-330, 009. [3] J. Huang, F. Jian, H. Wu, and H. Li, An improved level set method for vertebra CT image segmentation, Biomedical Engineering Online, 03 [4] Howlader N., Noone A.M., Krapcho M., Miller D., Bishop K., Altekruse S.F., Kosary C.L., Yu M., Ruhl J., Tatalovich Z., Mariotto A., Lewis D.R., Chen H.S., Feuer E.J., Cronin K.A., Brain and Other Nervous System Cancer, SEER Cancer Statistics Review, National Cancer Institute. Bethesda, pp. 975-03, 06. [5] Wang Xue-guang, Chen Shu-hong, An Improved Image Segmentation Algorithm Based on Two-Dimensional Otsu Method, International Journal of Engineering Research & Technology (IJERT), Vol., Issue 6, 0
[6] Pedro Rodrigues, J.L. Vilaca and Jaime Fonseca, Processing Application for Liver Tumour Segmentation, Life and Health Sciences Research Institute, School of Health Sciences, Minho University Braga, Portugal, 0. [7] Roopali R. Laddha and S.A. Ladhake, A Review on Brain Tumor Detection Using Segmentation and Threshold Operations, International Journal of computer Science and Information Technologies, vol. 5,, 04 [8] Y. Zhang and L. Wu, An MRI Brain Images Classifier Via Principal Component Analysis And Kernel Support Vector Machine, Information Science and Engineering, Southeast University, Nanjing, China, Vol 30, 0 [9] Rohini Paul Joseph, C. Senthil Singh, M.Manikandan, Brain Tumor Mri Image Segmentation And Detection In Image Processing, International Journal of Research in Engineering and Technology International Journal of Research in Engineering and Technology,04