Login New user?  
Applied Mathematics & Information Sciences
An International Journal


Volumes > Volume 17 > No. 5


An Enhancement Of Deep Learning Algorithm For Brain Tumor Segmentation Using Kernel Based CNN With M-SVM

PP: 1060-1082
R Thillaikkarasi, R A Sankaran, R Divya, R Vimali,
The brain tumor can be created by uncontrollable increase of abnormal cells in tissue of brain and it has two kinds of tumors: one is benign and another one is malignant tumor. The benign brain tumor does not affect the adjacent normal and healthy tissue but the malignant tumor can affect the neighboring tissues of brain that can lead to the death of person. An early detection of brain tumor can be required to protect the survival of patients. Usually, the brain tumor is detected using MRI scanning method. However, the radiologists are not providing the effective tumor segmentation in MRI image due to the irregular shape of tumors and position of tumor in the brain. Accurate brain tumor segmentation is needed to locate the tumor and it is used to give the correct treatment for a patient and it provides the key to the doctor who must execute the surgery for patient. In this paper, a novel deep learning algorithm (kernel based CNN) with M-SVM is presented to segment the tumor automatically and efficiently. This presented work contains several steps that are preprocessing, feature extraction, image classification and tumor segmentation of brain. The MRI image is smoothed and enhanced by Laplacian of Gaussian filtering method (LoG) with Contrast Limited Adaptive Histrogram Equalization (CLAHE) and feature can be extracted from it based on tumor shape position, shape and surface features in brain. Consequently, the image classification is done using M-SVM depending on the selected features From MRI image, the tumor is segmented with help of kernel based CNN method. Experimental results of proposed method can show that this presented technique can executes brain tumor segmentation accurately reaching almost 84% in evaluation with existing algorithms.

  Home   About us   News   Journals   Conferences Contact us