الفهرس | Only 14 pages are availabe for public view |
Abstract There is evidence that early detection of various diseases can improve the treatment and increase the survival rate of patients. The conventional method for diagnosing most of the existing diseases depends on human skills to recognize the occurrence of the convincing pattern. This age-old diagnosis method may subject to human mistake, imprecise diagnosis, time-consuming and labor intensive, and causes an unnecessary burden to radiologists. Moreover, by the right time of the diagnosis completed, it may already be at a critical stage. Recently, Computer Aided Diagnosis (CAD) and machine learning systems have been developed and functional in order to support specialists in determining the diagnosis decision process. However, medical diagnosis by most of the existing CAD systems depends on processing different types of digital images. This thesis presents an efficient CAD system for cancer diseases diagnosis by gene expression profiles of DNA microarray datasets. The proposed CAD system combines Intelligent Decision Support System (IDSS) and Multi-Agent (MA) system. The IDSS represents the backbone of the entire CAD system. It consists of two main phases; feature selection/reduction phase, and a classification phase. In the feature selection and reduction phase, eight diverse methods are developed. These methods include Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Correlated-based Feature selection (CFS), Information Gain (IG), Gain Ratio (GR), Relief-F, Chi-Square, and Support Vector Machine with Recursive Feature Elimination (SVM-RFE). While, in the classification phase, three evolutionary machine learning algorithms are employed, J48, Naïve Bayes (NB), Genetic Algorithm (GA). The IDSS performs the required cancer diseases classification Abstract It first receives a gene expression profiles dataset and then performs the feature selection and classification process. The feature selection is performed by using one of the eight approaches while the cancer classification is performed by using one of the three algorithms. Since there are eight feature selection methods and three classification methods, then the proposed CAD system allows 24 different IDSSs to be created and used for cancer diseases classification. On the other hand, the Multi-Agent (MA) system manages the entire operation of the CAD system. It first initializes several IDSSs (exactly 24 IDSSs) with the aid of mobile agents and then directs the generated IDSSs to run concurrently. Finally, a master agent selects the best classification, as the final report, based on the best classification accuracy returned from the 24 IDSSs. The proposed CAD system is implemented in JAVA, and evaluated by using eight microarray datasets included Breast cancer, Leukemia, Colon tumor, Central nervous system, Lung cancer Ontario, Lung cancer Michigan, Diffuse Large B-Cell Lymphoma and Prostate cancer. The main advantage of the proposed CAD system is that it classifies the cancer diseases accurately in a very short time. This is because cancer classification is done in parallel processing manner. Where, the MA system invokes 24 different IDSS to classify the diseases on the input dataset concurrently before taking a decision of the best classification result. |