Multivariate discriminant analysis of multiparametric brain MRI to differentiate high grade and low grade gliomas — A computer-aided diagnosis development study

Abstract

The aim of this study is to investigate the predictive capacity of multiparametric magnetic resonance imaging (MRI) findings using multivariate discriminant analysis. Preoperative clinical findings and multiparametric MRI, including diffusion weighted MR imaging, diffusion tensor imaging, perfusion MR imaging and MR spectroscopic imaging, were used as predictors to distinguish high grade from low grade gliomas. Principal component analysis was performed prior to discriminant analysis for dimensional reduction. Linear and quadratic discriminant analysis were performed and compared based on sensitivity and specificity analysis. The sensitivities of linear and quadratic discriminant analysis were 76.5% and 83.5%, respectively. Their specificities were 68.5% and 46.5%, respectively. Quadratic discriminant analysis provided a better discrimination than linear discriminant analysis for this dataset. This study is a model for a computer aided diagnosis system for glioma grading.