MRCT: Synthetic CT Imaging for Radiation Oncology Using Magnetic Resonance Images

Abstract

Purpose/Objective(s): MRI is an essential element of Radiation Oncology treatment planning. The ability of MRI data to yield synthetic CT images, specifically to define attenuation of skeletal features for dose calculation and image guidance support, was evaluated. Materials/Methods: Under a review board approved investigation, CT and various contrastMRI imageswere acquired of a dog. Imagevolumes acquired included 3D T1 gradient echo, T2 weighted, calculated water and fat (from a Dixon technique), and ultra short TE (UTE1, 70 microsecond echo time, followed by UTE2 at 3.7 ms). UTE1 images have signal intensities that are more strongly influenced by short T2 species, and show promise for separation of bone and air. Following intensity masking for separation of bone from air (using T1, UTE and water images), a classification algorithm (fuzzy c-means)was applied to themulti-spectralMRI volumes, yielding probability distributions for 5 classes (water, fat, brain, muscle, bone). Attenuationvalues were applied to each of the 5 classes and the resulting values were summed, yielding a synthetic CT (MRCT) volume. The actual CT volume was rigidly aligned with the MRI volumes (with specific focus on the skull). Regions of interest were defined based on increasing intensity thresholds on the CT scan (from400-1300HU in 300HUsteps), representing regions of increasing bone density. Intensities within these regions were evaluated on the MRCT. Intensities of MRI volumes of each contrast, as well as probabilities of class membership, were similarly evaluated within the same regions of interest. Results: Each of the input imaging sequences showed a trend of decreasing average signal intensity for regions generated via increasing CT thresholds, as expected due to the loss of signal in tissues with bone. All tissue classes except for bone decreased in membership probability in regions of increasing CT intensity, with the exception of the bone class, which increased with increasing CT value. Intensities within regions of varying bone density in the MRCT volume correlated strongly with those of the actual CT (R of 0.994). Digitally reconstructed radiographs from the actual CT volume were successfully aligned to matching projections through MRCT volumes without bias. Conclusions: The information present in MRI is sufficient to support identification and spatial mapping of skeletal features and classification can properly assign attenuation values to the heterogeneous skeletal tissues in the head. Further work will focus on assessing the accuracy of mapped attenuation of other tissues in the body and on treatment plan and IGRT comparisons using actual and synthetic CT data. Author Disclosure: Y. Cao: E. Research Grant; NIH. S. Hsu: None. A. Flammang: A. Employee; Siemens Corporate Research. J. Balter: E. Research Grant; NIH.