Max-AUC Feature Selection in Computer-Aided Detection of Polyps in CT Colonography

February 26, 2014

J.-W. Xu and K. Suzuki

Max-AUC Feature Selection in Computer-Aided Detection of Polyps in CT Colonography

A feature selection method based on a sequential forward floating selection procedure is proposed to improve a non-linear SVM (Support Vector Machine) classifier’s performance in detection of polyps in CTC (Colongraphy). The proposed method selected the most relevant features that would maximize the AUC (Area Under the receiver-operating characteristic (RoC) Curve). The method was compared against the popular stepwise feature selection method based on Wilk’s lambda for a colonic-polyp database (25 polyps and 2,624 non-polyps). Significant improvement in the classifier performance has been found with the proposed method, where 4.1 and 6.5 False Positive (FP) rates per patient were found when using two SVM classifiers trained with the features selected by the proposed method and 18.0 FP per patient was found when stepwise feature selection was used.

Read more at IEEE Xplore

Tags: Colonic polyps, computer-aided detection, feature selection, support vector machines

 

Login

New Here? Sign Up

Looking for increased exposure in the field of biomedical engineering? EMBS offers journals, conferences and a community for biomedical engineers. Membership includes PULSE Magazine.

Join EMBS