New Reliable Cancer Diagnosis Method Using Boosting and Projective Adaptive Resonance Theory

For adequate treatment of patients, it is important that the accurate and reliable algorithm is developed for construction of diagnosis system. Gene expression data from DNA microarray are individualized and are useful in the diagnosis and prognosis of diseases. However, to conduct analysis, it is necessary to select marker genes that are strongly related to diagnosis or prognosis of disease because the performance of classification analysis can decline due to such large quantities of data. Feature selection has been performed in order to screen candidate genes for modeling. There are two types of approaches-wrapper approach and filter approach. In the former approach, features (genes) are selected as a part of mining algorithms. On the other hand, in the filter approach, features are selected by filtering methods prior to the application of mining algorithms. In the previous study, we developed boosted fuzzy classifier with SWEEP operator (BFCS) method [1] as a wrapper approach, and projective adaptive resonance theory (PART) [2] as a filter approach. In the present study, we combined PART and BFCS and then, we applied this method to gene expression profile data of central nervous system (CNS) tumor. The results obtained by modeling were evaluated by comparison with the combination method of various modeling and filtering algorithms.