Medical Physics Seminar – Monday, April 1, 2013
Support Vector Machine Tissue Classification of Multiparametric MRI Tumor Data
J. Gabe Heredia (student of Dr. M. Elizabeth Meyerand)
Research Assistant, Department of Medical Physics, UW-School of Medicine & Public Health, Madison, WI - USA –
Glioblastoma Multiforme (GBM) is the most prevalent and aggressive type of malignant brain tumor affecting humans. The median survival ranges from 4.5 months with no treatment, to 15 months with resection, radiation therapy, and chemotherapy. Recurrence occurs in nearly 95% of patients, and this is due in part to the highly infiltrative and heterogeneous nature of GBM. The ability to better characterize the heterogeneous makeup of these tumors could allow for more aggressive treatment. Our study consisted of collecting multiple MR images from patients with confirmed GBM, and repeating scans every 3 months. The MRI data consisted of T2, post-contrast T1, perfusion based rCBV and permeability maps, diffusion (ADC), and a carbogen-based hypoxia map. These images were collected for each patient and coregistered to provide the data matrix. They were then analyzed by a radiologist and assigned tissue labels to complete the input matrix for the support vector machine classifier (SVM). This presentation will discuss the ability of a support vector machine classifier to disambiguate healthy tissue from, tumor, necrosis, edema, and cyst, based on the multiparametric MRI data in patients with GBM.