NIMG-40. RADIOMICS-BASED DIFFERENTIATION OF GLIOBLASTOMA AND METASTATIC DISEASE: DOES DIFFERENCE IN TYPE OF T1-CONTRAST ENHANCED SEQUENCE MATTER?

Abstract

Abstract BACKGROUND To evaluate the radiomics-based model performance for differentiation between glioblastoma (GB) and intracranial metastatic disease (IMD) using magnetization prepared rapid gradient echo (MPRAGE) and volumetric interpolated breath-hold examination (VIBE) T1-contrast enhanced sequences. MATERIALS AND METHODS T1-CE MPRAGE and VIBE sequences acquired in 108 patients (31 GBs and 77 IMD) during the same MRI session were retrospectively evaluated. Post standardized image pre-processing and segmentation, radiomics features were extracted from necrotic and enhancing tumor components. A total of 90 machine learning (ML) pipelines were evaluated using a five-fold cross-validation. Performance was measured by mean AUC, Log-loss, and Brier scores. Pearson correlation analysis of radiomics features from tumor subcomponents was also performed. RESULTS The mean AUC across the top-ten pipelines varied between 0.890-0.851 with T1-CE MPRAGE and 0.907-0.869 with T1-CE VIBE sequence. Top performing models for the MPRAGE sequence commonly used support vector machines, while those for VIBE sequence used either support vector machines or random forest. Common feature reduction methods for top-performing models included linear combination filter and least absolute shrinkage and selection operator (LASSO) for both sequences. Only three of the top ten models used the same combination of feature reduction-ML algorithm pipeline. A feature-wise comparison showed that the radiomic features between sequences were strongly correlated, with the highest correlation for shape-based features. CONCLUSION Radiomics features derived from T1-CE MPRAGE and VIBE sequences may have similar overall classification performance for differentiating GB from IMD, albeit often using different feature selection-ML algorithm pipelines.