A Baseline for Predicting Glioblastoma Patient Survival Time with Classical Statistical Models and Primitive Features Ignoring Image Information

摘要

Gliomas are the most prevalent primary malignant brain tumors in adults. Until now an accurate and reliable method to predict patient survival time based on medical imaging and meta-information has not been developed [3]. Therefore, the survival time prediction task was introduced to the Multimodal Brain Tumor Segmentation Challenge (BraTS) to facilitate research in survival time prediction. Here we present our submissions to the BraTS survival challenge based on classical statistical models to which we feed the provided metadata as features. We intentionally ignore the available image information to explore how patient survival can be predicted purely by metadata. We achieve our best accuracy on the validation set using a simple median regression model taking only patient age into account. We suggest using our model as a baseline to benchmark the added predictive value of sophisticated features for survival time prediction.