Advanced quantitative MRI radiomics features for recurrence prediction in glioblastoma multiform patients

Introduction: Advanced quantitative information such as radiomics features derived from magnetic resonance (MR) image may be useful for outcome prediction, prognostic models or response biomarkers in Glioblastoma (GBM). The main aim of this study was to evaluate MRI radiomics features for recurrence prediction in glioblastoma multiform.   Materials and Methods: 86 patients with recurrent GBM who underwent MRI were subjected to this study. The axial T1-weighted contrast-enhanced and axial T2-weighted FLAIR images were included for analysis. All images were preprocessed by different bin width (32, 64 and 128). For each lesion we manually segmented Active, Necrosis and whole Tumor region in T1-CE and Edema region in T2-FLAIR. 105 quantitative 3D features and texture based on intensity histograms (IH), gray level run-length (GLRLM), gray level co-occurrence (GLCM), gray level size-zone texture matrices (GLSZM), neighborhood-difference matrices (NDM), and geometric features were extracted from the 3D-tumor volumes of each segment. Random Forest (RF) machine learning with 10-fold cross validation was used to recurrence prediction in GBM. Results: Area under ROC curve (AUC) as an assessment index on RF with bin width of 32, 64 and 128 achieved in Active (0.616, 0.586, 0.509), Necrosis (0.521, 0.521, 0.545), whole Tumor (0.639, 0.602, 0.547) and Edema regions (0.629, 0.669, 0.621), respectively. Conclusion: The main purpose of this assay was to assess the power of MRI radiomics features in GBM patients for recurrence prediction. The proposed method can effectively predict recurrence in GBM by application of advanced MRI quantitative radiomics features and machine learning.