Radiogenomics of Glioblastoma: Identification of Radiomics associated with Molecular Subtypes

• Main Authors: Wijethilake, Navodini, Islam, Mobarakol, Meedeniya, Dulani, Chitraranjan, Charith, Perera, Indika, Ren, Hongliang
• Format: Journal Article
• Language: English
• Published: 27-10-2020
• Subjects: Computer Science - Artificial Intelligence; Computer Science - Learning; Quantitative Biology - Quantitative Methods
• DOI: 10.48550/arxiv.2010.14068

Glioblastoma is the most malignant type of central nervous system tumor with
GBM subtypes cleaved based on molecular level gene alterations. These
alterations are also happened to affect the histology. Thus, it can cause
visible changes in images, such as enhancement and edema development. In this
study, we extract intensity, volume, and texture features from the tumor
subregions to identify the correlations with gene expression features and
overall survival. Consequently, we utilize the radiomics to find associations
with the subtypes of glioblastoma. Accordingly, the fractal dimensions of the
whole tumor, tumor core, and necrosis regions show a significant difference
between the Proneural, Classical and Mesenchymal subtypes. Additionally, the
subtypes of GBM are predicted with an average accuracy of 79% utilizing
radiomics and accuracy over 90% utilizing gene expression profiles.