الفهرس 
Medical Imaging Techniques for Human BrainOnly 14 pages are availabe for public view
 |  | from | 137 |  |  |
| | | from | 137 | | |
Abstract
Machine learning became very important in extracting meaningful relationships and making accurate prediction in many fields. In the area of processing the brain images, Computer Aided-Diagnosis (CAD) systems are basically relied on different machine learning techniques in all its stages to implement a system that can help the radiologists by providing a second opinion that can assist in detection and diagnosis of brain tumors based on imaging techniques that are widely used in clinical care. Magnetic resonance imaging (MRI) is an imaging technique that plays a vital role in detection and diagnosis of brain tumors in both research and clinical care for providing a detailed information about the brain structure and its soft tissues.
This study concerned with developing a CAD system that can process the brain MR images for detection and diagnosis of different brain tumors using several machine learning techniques. In this study, two types of systems are implemented. The first type is to differential diagnose of 3 types of malignant brain tumors (i.e., glioblastoma, sarcoma and metastatic bronchogenic carcinoma) from normal brain subjects using brain MRIs from a real online dataset of brain MRIs. However, the second type is to differential diagnose of cognitive normal (CN) brain from Alzheimer’s disease (AD) brain subjects using brain MRIs from two real online datasets of brain MRIs.
For the first type of CAD system, three CAD systems with several models are presented in this study through the chapters. The three CAD systems included three stages: segmentation, feature extraction and selection and classification. K-means and Fuzzy C-means are two segmentation techniques that have been used separately to segment the input brain MRIs from the dataset used. Gray level co-occurrence matrix (GLCM) and Discrete Wavelet Transform (DWT) integrated with Principal Component Analysis (PCA) are also two techniques for feature extraction and selection that have been used separately. The three CAD system models are combination of these techniques in the first two stages with a selected
V
classifier for the final stage. In the last stage, for the classification three different supervised classifiers are used for each system: (i) hybrid Support Vector Machine (SVM) which is integration of Linear-SVM for differentiating between normal and abnormal brain subjects and Multi-SVM classifier for identifying the tumor type in the abnormal brain subjects, (ii) SMO-SVM which is sequential minimal optimization (SMO) algorithm for training Support Vector Machine (SVM) and (iii) Deep Neural Network (DNN). The performance of the three classifier is evaluated using different measures giving high average classification rates and precision that proved their efficiency and reliability.
The second type of CAD systems presented in this study is a CAD system for differential diagnose of CN brain from AD brain subjects based on Linear Discriminate Analysis (LDA) classifier. The developed system includes two stages: feature extraction and selection and classification stage. Extracting the features from the input brain MRIs of the two datasets used is done for each dataset separately using DWT integrated with PCA for reducing the number of features to avoid classification complications and reduce the computation time and costs. The system presented is tested using two different datasets obtained from online datasets of real human brain MRIs. The performance of the presented system proved its efficiency and reliability in the problem which it is used for according to different performance measures.