Overall survival prediction using conventional MRI features

Abstract

Gliomas are common primary brain malignancies. The sub-regions of gliomas are depicted by MRI scans, reflecting varying biological properties. These properties have effect on the diagnosis of neurosurgeons on whether or what kind of resection should be done. The survival days after gross total resection is also of great concern. In this paper, we propose a semi-auto method for segmentation, and extract features from slices of MRI scans, including conventional MRI features and clinical features. 13 features of a subject are selected finally and a support vector regression is used to fit with the training data.