Feasibility of Generating Synthetic ED and CT Images From Conventional MR Images for Brain Radiation Treatment Planning and Delivery

Abstract

Magnetic resonance imaging (MRI) provides soft tissue and functional contrast that may improve target delineation in radiation treatment planning. However, in order to establish MRI as the primary imaging modality in radiation therapy, limitations of lack of extractable electron density (ED) information and incompatibility with existing CT-based image guided radiation therapy (IGRT) technologies must be overcome. The purpose of this study was to determine the feasibility of generating synthetic ED and synthetic CT images directly from conventional MRI images for use in brain radiation treatment planning and delivery. A retrospective study was performed using conventional MRI images acquired on three brain tumor patients. Pre-contrast T1 and T2 images were segmented into air and skin, scalp, skull, brain, cerebrospinal fluid, and eye meshes using FSL and AFNI software. Relative ED values, obtained from ICRU Report 46, were applied to each segmented structure. The segmented structures were then recombined in to form synthetic ED images. Synthetic CT images were generated from the synthetic ED images by mapping ED to Hounsfield Units using the standard CT-ED conversion table from a clinical radiation treatment planning system. The synthetic CT images were then transferred to a radiation treatment planning system and sample whole brain radiation plans were generated without and with heterogeneity correction. Finally, the synthetic CT images were transferred to a clinical IGRT workstation. In general, the combined use of multiple MR images improved segmentation robustness. However, skull segmentation proved to be difficult in inferior cranial regions. Differences in dose distributions in plans generated without and with heterogeneity correction were apparent, indicating that the planning system was able to successfully convert the synthetic CT images into electron density. Digitally reconstructed radiographs (DRRs) were successfully generated and the synthetic CT images were able to be successfully loaded as “planning CT” images for IGRT registration during radiation treatment. Generation of synthetic ED and CT images directly from conventional MRI images for brain radiation treatment planning and delivery is feasible. However, regions of complex anatomy and/or poor image contrast may hinder segmentation and thus, dose distribution, accuracy. Future work will examine MRI relaxometry-based segmentation methods, which may improve robustness in the presence of image non-uniformities, and compare radiation treatment plans generated using synthetic CT images to those generated using conventional CT images.