Magnetic Resonance Texture Analysis: Optimal Feature Selection in Classifying Child Brain Tumors

Textural feature based classification has shown that magnetic resonance images can characterize histological brain tumor types. Feature selection is an important process to acquire a robust textural feature subset and enhance classification rate. This work investigates two different feature selection techniques; principal component analysis (PCA), and the combination of max-relevance and min-redundancy (mRMR) and feedforward selection. We validated these techniques based on a multi-center dataset of pediatric brain tumor types; medulloblastoma, pilocytic astrocytoma and ependymoma, and investigated the accuracy of tumor classification, based on textural features of diffusion and conventional MR images.