Meeting Banner
Abstract #0180

Feature Analysis in SVM-Based Classification of Gliomas

Frank G. Zoellner1, Kyrre E. Emblem2,3, Lothar R. Schad1

1Computer Assisted Clinical Medicine, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany; 2Department of Radiology, MGH-HST A. A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital & Harvard Medical School, Boston, MA, USA; 3The Interventional Center, Oslo University Hospital, Oslo, Norway

DSC-MRI is a method of choice to differentiate high-grade from low-grade gliomas. Recently, support vector machine (SVM) learning have been introduced as means to prospectively characterize gliomas based on the rCBV histograms. In our study, we have assessed the diagnostic accuracy of the different histogram features used in the SVM analysis (peak height, skewness, etc). By using correlation analysis to reduce 95% of the feature information, a classification accuracy of 88.1% was yielded. Our results suggest that a careful examination of the features in SVM based glioma grading could reduce the number of features substantially, thereby improving the effectiveness of the SVM analysis while maintaining a good classification score.