A mixture classifier for computer aided diagnosis of polyp malignancy for CT colonography

Abstract

Feature classification is an important part in computer-aided diagnosis of suspicious lesions. Currently there are many classifiers available, e.g., support vector machine (SVM), random forest (RF) and linear discriminant analysis (LDA). However, each of the classifiers has advantages and drawbacks and may show good performance in some cases and cannot show good classification in some other cases. It has been observed that different classifiers have different performances. This observation inspires us to explore a new classifier that can overcome the limitations of each single classifier while retaining the advantages of each single classifier. In this paper, we explored two “mixture” classifiers, one is the combination of two among the SVM, RF and LDA, and the other is the combination of all three. The performances of the two mixture classifiers were compared with respect to each individual, i.e., SVM, RF and LDA using a colon polyp database, including 116 neoplastic lesions and 37 hyperplastic lesions. The performances were quantitative measured by the area under the curve (AUC) of the Receiver Operating Characteristics. The results show that the “mixture” classifiers can have a better performance than each individual classifier, respectively. The running time of the mixture classifiers is dominated by the time of the SVM.