Feasibility of MR-Generated Synthetic CT Images for Planning Stereotactic Radiosurgery of Brain Metastases

Abstract

Frameless Linac-based stereotactic radiosurgery (SRS) treatment planning typically involves spatial registration of MR and CT images. A solely MR-guided workflow would eliminate the burdens of CT scanning and associated time as well as alignment errors, but would require dose calculations and patient positioning based on MR-derived information. The feasibility of MR-guided SRS of brain metastases was investigated. Specifically, it was hypothesized that synthetic CT images (MRCTs) could be used to generate plans and reference images for positioning that would not differ from SRS using a traditional CT-guided technique. MR and CT images were acquired for 19 patients undergoing SRS for brain metastases. Following rigid image alignment, PTV and OAR contours were delineated. MRCT image volumes were generated using a voxel-based probabilistic classification technique and aligned rigidly to their respective CT scans. The CT structures were transferred onto the MRCT volumes. CT-based plans were created with the intent to deliver 99.5% of the prescription dose to the PTV and avoid excess doses to the OARs. The fluences from the CT plans then were used for dose calculations on a “maskless” CT plan (in which the attenuation of the immobilization device was removed) and MRCT images. PTV D99.5 and OAR D0.1 cc values were compared. OARs (brainstem, optic nerves/chiasm, eyes, lenses, cochleae) typically defined on CT were drawn on the source MR (T1- and T2-weighted image volumes) images and compared to the original CT volumes. Cone beam CT (CBCT) volumes used for positioning at treatment were aligned to the MRCT and CT volumes, and resulting transforms compared. A total of 43 PTVs were studied with a mean prescription dose of 21.1 Gy (range 12 to 30 Gy). The mean difference in PTV D99.5 between the CT versus maskless CT was 0.0 % (range -0.0 to 0.1 %) and CT versus MRCT was 0.0 % (range -0.1 to 0.0 %) of the prescription doses. Dose metrics for a total of 185 OARs were calculated. OAR D0.1 cc difference between the CT versus maskless CT was 0.0 Gy (range -0.1 to 0.0 Gy) and CT versus MRCT was 0.0 Gy (range -0.2 to 0.7 Gy). For OAR volumes larger than 1 cc (n=64), the mean Dice coefficient between CT and MR structures was 0.82 (range 0.17 to 0.93). Mean differences in alignment between the CBCT volumes and MRCT and CT volumes resulted in isocenter positioning differences of 0.0 mm (range -0.2 to 0.1 mm), 0.0 mm (range -0.5 to 0.4 mm), and 0.1 mm (range -0.2 to 0.5 mm) in the LR, AP, and CC directions. The accuracy of delineation, calculated doses, and patient positioning indicate that MR-guided frameless SRS is feasible. Future investigations will explore generation of reference image volumes for alignment that combine MR and synthesized CT contrasts. Supported by NIH R01EB016079.