Improving classifications of brain tumor tissue with Sparse Dictionary Learning based analysis of dynamic susceptibility contrast enhanced MRI data

We analyzed the DSC MRI signals based on patterns of descriptive DSE-MR parameters by using Sparse Dictionary Learning (SDL) coding method. We successfully decomposed DSC MRI signals into linear combinations of multiple components based on sparse representation of DSC MRI signals in the tumor region of tumor core and peritumoral edema which might be represent multiple heterogeneity component in brain tumors. Assessment of diagnostic performance of SVM classification after cross validation revealed that the combination of conventional DSC temporal characteristics and dictionary learning based DSC temporal features would result in the best classification accuracy between tumor core and peritumoral edema (with total diagnostic accuracy of 77%, AUC 0.78).