Detection of malignant pleural effusion based on features of cell

摘要

In this paper, we used automated method to detect pleural effusion automatically based on the feature selection of tumor cell. Two kinds of feature selection method, filter approach and wrapper approach are used. Our experimental results show that the wrapper method using GA search engine can yields a good classification result and filter method is not fit for the detection of pleural effusion. Also, we compared the performance of classifiers with feature selection. The three classifiers are MLP, SVM with polynomial kernel and SVM with RBF kernel. The results show that SVM with polynomial kernel outperform the other two classifiers in most cases in the diagnosis of malignant effusions. The accuracy of SVM-Poly classifier with GA wrapper feature selection is very high. The success of the SVM with wrapper feature selection in the effusion data suggest they may be even more promising for their applications in progression prediction and for automatic diagnosis of other disease based on the feature selection of cells.