الفهرس 
INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 151 |  |  |
| | | from | 151 | | |
Abstract
A brain tumor is one of the most devastating diseases. Early detection of brain tumor is a life-saving act. Magnetic Resonance Imaging (MRI) is one of the main techniques to detect brain tumor for diagnosis and treatment. Although there are numerous methods for brain tumor segmentation, automatic and exact segmentation still confronted with some problems and remain one of the most challenging tasks in medical data processing.
This work presents a complete algorithm to classify MRI scans between two types of class (Normal scan class and Abnormal scan class). Then segment the tumor from the abnormal scans. the classification is carried out by using Discrete Wavelet Transform )DWT( for feature extraction, Gray Level Co-occurrence Matrix )GLCM( and Principal Component Analysis )PCA( for selecting the valuable features with additional morphological based features to classify between the two classes. this fixed technique of extracting , reducing and selecting features is fed to three types of classifiers which are Kernel Support Vector Machine (KSVM) with Radial basis Function (RBF) kernel, Linear Discriminant Analysis (LDA) and Artificial Neural network (ANN), which are tested next on a large database with accuracy of 93.06%, 97.45% and 98.9% respectively. The detection of the brain tumor region from the abnormal scan images is carried out by using a hybrid method that is based on morphological operations, Filtering and Histogram Processing on images.
The proposed method consists of two main flow charts which are segmentation flow chart and classification flow chart. The reason of splitting the system flow chart into two flow charts is to test the segmentation flow chart separately before using it in the system. The segmentation flow chart is then used as a part of the classification flow chart when the abnormal scan classification results are fed to the segmentation part. The segmentation flow chart is tested on a different database that provides the tumor result. This result image is then used to evaluate the output result of segmentation. The evaluation is conducted by calculating true positive ratio (sensitivity) (TPR) - true negative ratio (specificity) (TNR) - predictive value positive (PVP) and accuracy (A).
The best performance of classification is to use artificial neural network as pattern recognition feed forward multilayer perceptron, Rectified linear unit (ReLU) activation function in the hidden layers, an output layer with Softmax classification function. This is used to perform Multinomial Logistic Classification. Bayesian Regularization Back-propagation training technique is used to train the neural network which updates the weight and bias values according to Levenberg-Marquardt optimization to minimizes a combination of squared errors and weights, and then determines the correct combination in order to produce a network that generalizes well.
This thesis consists of (6) chapters as following:
Chapter 1 presented the definitions (Brain, brain functions, brain tumor, MRI, MRI scan …etc) then illustrated the problem, motivation, challenges, research objectives and contributions of this research.
Chapter 2 illustrates the related work for MRI scan classification and brain tumor extraction.
Chapter 3 Explains tumor extraction Flow chart and its procedures with the methods involved such as histogram processing, noise types and its affection on MRI images and how to recognize and eliminate it, segmentation and morphological operations that are used in the segmentation flow chart.
Chapter 4 provides Classification Flow chart and the methods used within it.
Chapter 5 presents the results from conducting the proposed method on the database, and comparing the results with the results from recent researchers work.
Chapter 6 concludes results from tumor extraction and MRI scan classification parts and future work.