Comparison of Single MRI vs. Multiparametric MRI in GBM Image Data.

Purpose/Objective(s) MRI is the standard imaging tool for diagnosis, monitoring and response evaluation for glioblastoma multiforme (GBM). The diagnostic image sequences typically include T1-weighted contrast-enhanced (T1-C), T1-weighted (T1), T2-fluid-attenuated inversion recovery (T2-Flair), standard T2-weighted (T2), and apparent diffusion coefficient (ADC) through diffusion-weighted imaging. Multiparametric MRI (mpMRI) integrates these image data to define spatially distinct regions with similar physiologies as habitats. This study tests the hypothesis of whether mpMRI, single MRI sequence, or combination sequences could improve prediction of GBM recurrence. Materials/Methods A study cohort of 119 patients with GBM was used for evaluation. The primary outcome was recurrence which was defined as low risk if recurrence time was above the median recurrence time and otherwise as high-risk. For mpMRI, analysis was limited to 2 image sequence combinations from the above 5 image sequences, resulting in 10 sets of 2-image sequences for habitat features (e.g., T1C and T2-Flair, T1C and ADC, etc.). Analysis for single MRI sequences was divided into 14 subgroups by different filters (e.g., Wavelet, neighborhood, and intensity). Subgroups from two image sequences from same filter were combined for analysis, as well as the habitat features data from the same image sequences. Random forest method was used to evaluate performance in each group based on average test accuracy rate with a random split of 84 in training set and 35 in test set for 10 times of shuffling. Results The Wavelet subgroup performed better than other subgroups. Most of the average accuracy rates for two image sequences combinations in this subgroup were more than 75%, with the highest rate of 79% in the combination of T1 and ADC. The highest accuracy rate for the single MRI subgroup is 80% in the T2F's Wavelet subgroup. The mpMRI had an average test accuracy rate of 66%. Conclusion This exploratory analysis indicates that the Wavelet filter could be an informative image marker in predicting GBM recurrence. Future large study cohorts are needed to verify the results.