Purpose: The purpose of this study is to generate relaxation maps from clinical MRI scans by leveraging prior knowledge. Furthermore, we analyze joint histograms of these maps (T1 pre contrast, T1 post contrast, and T2 relaxation maps) to investigate glioma tumor heterogeneity.
Methods: This study uses 50 patients from the multi-modal brain tumor image segmentation benchmark (BRATS) dataset, which contains four tumor sub-type volumes (necrotic core, edema, non-enhancing and enhancing tumor), T1-weighted (T1-w), T2-weighted (T2-w), and T1-weighted post contrast (T1c-w) MR images. By employing a novel method, we generate relaxation maps (T1, T2, and T1c maps) by leveraging relaxation values of white matter (WM) and cerebral spinal fluid (CSF) from the literature and calibrating the clinical MRIs based on these values. After the calibration process, the gray matter (GM) T1 and T1c relaxation values are automatically assessed for quality control purposes. 3D joint histograms (JH) were then created using all voxels within the tumor and each tumor sub-type. We calculated the JH center of masses, which are defined in the T1-T2-T1c relaxation coordinate system, for each sub-type, and quantitatively evaluated the relationship of these center of masses using a correlation matrix across all patients.
Results: Relaxation maps for 6 patients were deemed unacceptable for further analysis using the developed GM quality control metric. Multiple subtypes’ center of mass correlations across the remaining 44 patients showed correlation values of r>0.5, corresponding to p<0.0005. The highest correlation is 0.894 between the non-enhancing T1 and the non-enhancing T1c, meaning that the relationship between these relaxation values is most consistent across all patients.
Conclusion: We present a novel method capable of estimating relaxation maps for any set of T1-w, T2-w, and T1c-w clinical MRIs, and believe these maps can be valuable tools capable of characterizing changes in tumor heterogeneity over time.