MACHINE LEARNING BASED OVERALL SURVIVAL PREDICTION OF GLIOBLASTOMA MULTIFORME PATIENTS USING MAGNETIC RESONANCE IMAGE DERIVED FEATURES

Abstract

Glioblastoma multiforme (GBM) is a rapidly growing tumor associated with poor prognosis. The ability to predict overall survival (OS) of GBM patients enhances their surgical and treatment planning. In this work, OS prediction was performed using machine learning with a comprehensive set of features comprising of multi-channel magnetic resonance image (MRI) derived texture features, tumor shape and volumetric features by two different approaches: (a) OS group classification and (b) continuous value estimation via regression. In both approaches, high prediction accuracy was obtained. The tumor shape features used in this work have not been analyzed for OS prognosis in GBM patients previously. Their effectiveness was evaluated using Cox regression and Kaplan-Meir survival analysis. The imaging biomarkers used in this study were extracted from routinely acquired multi-channel MR images, making the translation of this work into a clinical setup convenient.