Evaluation of dosimetric uncertainty caused by MR geometric distortion in MRI‐based liver SBRT treatment planning

Abstract Purpose MRI‐based treatment planning is a promising technique for liver stereotactic‐body radiation therapy (SBRT) treatment planning to improve target volume delineation and reduce radiation dose to normal tissues. MR geometric distortion, however, is a source of potential error in MRI‐based treatment planning. The aim of this study is to investigate dosimetric uncertainties caused by MRI geometric distortion in MRI‐based treatment planning for liver SBRT. Materials and Methods The study was conducted using computer simulations. 3D MR geometric distortion was simulated using measured data in the literature. Planning MR images with distortions were generated by integrating the simulated 3D MR geometric distortion onto planning CT images. MRI‐based treatment plans were then generated on the planning MR images with two dose calculation methods: (1) using original CT numbers; and (2) using organ‐specific assigned CT numbers. Dosimetric uncertainties of various dose‐volume‐histogram parameters were determined as their differences between the simulated MRI‐based plans and the original clinical CT‐based plans for five liver SBRT cases. Results The average simulated distortion for the five liver SBRT cases was 2.77 mm. In the case of using original CT numbers for dose calculation, the average dose uncertainties for target volumes and critical structures were <0.5 Gy, and the average target volume percentage at prescription dose uncertainties was 0.97%. In the case of using assigned CT numbers, the average dose uncertainties for target volumes and critical structures were <1.0 Gy, and the average target volume percentage at prescription dose uncertainties was 2.02%. Conclusions Dosimetric uncertainties caused by MR geometric distortion in MRI‐based liver SBRT treatment planning was generally small (<1 Gy) when the distortion is 3 mm.


| INTRODUCTION
The American Cancer Society estimates that in 2018, about 42 220 adults (30 610 in men and 11 610 in women) in the United States will be diagnosed with primary liver cancer. 1 Liver is also a common site of metastases. 2,3 Nearly 70-90% of liver metastases cannot be resected through surgery. 3 Stereotactic body radiotherapy (SBRT) has been shown to improve the local control rate of liver cancer.
Different from the conventional radiation therapy which uses low fractional dose of~2 Gy/fx, SBRT has a substantially greater cell-killing effect using very high fractional dose of 10-20 Gy/fx, 4,5 leading to the excellent local tumor control rates of >90% if adequate radiation dose is delivered. 6 Sharp dose fall off outside the target volume of SBRT requires precise contouring of target volume and organs at risk (OARs). 7,8 Advanced imaging techniques, such as 4D-CT, are commonly used for precise tumor volume contouring in SBRT.
Current liver SBRT technique is CT-based. However, CT is known to have low soft tissue contrast, and thus inability to accurately determine tumor volume and tumor motion in the abdomen. [9][10][11][12] In current liver SBRT treatment planning, MRI is often fused to CT to assist target volume delineation. This approach is not ideal as the registration between CT and MRI is prone to errors, and large safety margin is often needed to compensate for this uncertainty, 13 which will increase the radiation dose to OARs. Therefore, current CT-based liver SBRT treatment planning is ineffective and inefficient.
It requires multiple imaging scans (CT, multiple MRI, etc.) and additional planning efforts (CT-MRI registration, contour transfers, etc.), which increases the planning time, cost, and associated uncertainties.
There is a clear need for improved liver SBRT technology.
Compared to CT, MRI has many significant advantages for radiotherapy planning, including superior tumor and soft-tissue contrast, flexible imaging orientation, freedom from radiation exposure, and real-time imaging. MRI-based treatment planning is an emerging technique that can potentially improve tumor volume accuracy and dosimetry as compared to CT-based planning for certain cases. MRIbased treatment planning has been developed in brain, head and neck, and prostate. Paradis et al. 14 proposed to generate synthetic CT images by segmenting brain tissues in MR images based on probabilistic classification using fuzzy c-means clustering and assigning corresponding weighed CT number to each voxel. Dosimetric comparison was performed between CT-based and MRI-based brain volumetricmodulated radiation therapy treatment planning. Hsu et al. 15 investigated probabilistic tissue classification based on fuzzy c-means clustering for soft tissue segmentation in head and neck, and performed CT number assignment according to the International Commission on Radiation Units and Measurements (ICRU) Report 46. Chen et al. 16 illustrated MRI-based treatment planning for prostate intensity-modulated radiation therapy (IMRT) through Atlas registration.
MRI-based treatment planning is a promising technique for liver SBRT, improving target volume delineation accuracy and reducing radiation dose to the OARs. However, MRI geometric distortion is a known important concern in MRI-based treatment planning 17,18 and may cause dosimetric uncertainties, which is especially critical for liver SBRT due to its hypofractionation. It is therefore the goal of this study to quantitatively evaluate the dosimetric uncertainties caused by MR geometric distortion in MRI-based liver SBRT treatment planning. We performed computer simulation studies based on measured MRI distortion data and clinical liver SBRT plans to evaluate the dosimetric effects of various scenarios of MRI distortion.
2 | MATERIALS AND METHODS 2.A | 3D MRI distortion simulation based on measured distortion data 3D MRI distortion was simulated based on sparsely measured MRI distortion data reported in the literature. 17,18 All MR images are expected to be distortion corrected using vendor's correction algorithms prior to any radiation therapy application. The simulated distortions consider only the system related distortions mainly caused by the inhomogeneities of main magnetic field B 0 and nonlinearities of the gradient coils. And they selected a relatively high receiver bandwidth to acquire data to decreases the signal-to-noise ratio, while reducing the distortion caused by susceptibility and chemical shift. Therefore, the distortion measurement is mainly from the system related distortion caused by B 0 inhomogeneities and gradient coil nonlinearities. Therefore, the simulated MRI distortion here refers to the residual MRI distortion, which is of concerned in MRIbased treatment planning. Table 1 summarizes the sparsely measured residual MRI distortion data used in their study. 17,18 These data were measured on phantoms and were the averaged values of multiple investigated MRI sequences, including 2D Axial T1 FSE, 2D Axial T2 FRFSE, 3D CUBE T1, 3D CUBE T2, and 3D T1 3D SPGR.
The in-plane MRI distortions were calculated as the average of the distortion along the X and Y directions, and the through-plane distortions were calculated along the Z direction.
The simulated in-plane and through-plane MRI distortions, Disin(r1 ) and Dis thr (r 2 ) respectively, was obtained by fitting the sparsely measured in-plane and through-plane MRI distortions (Table 1) using a two-term exponential fitting model as shown in Eq. (1a) and (1b), where r 1 is the radial distance from each pixel at in-plane image to the corresponding in-plane image center (x = 0, y = 0) and r 2 is the distance from each slice location to the central slice location (z = 0).
Then the radial distance r from each voxel to the scanning center (x = 0, y = 0, z = 0) and the 3D MRI distortion Dis sim (r) was then calculated as Eq. (2a) and (2b): Finally, a synthetic planning image dataset with simulated distortion was generated by deforming the original planning image dataset using the above-determined 3D MRI distortion.

2.B | 3D MRI distortion simulation based on body shrinkage
It was found from the above simulation study that meaningful anatomical changes due to MRI distortion mainly occur near the body surface. 17,18 To efficiently evaluate the dosimetric effects due to different distortion magnitudes, we used a simple "body shrinkage" method to generate the distorted planning images by shrinking the body contour with a preset value (2, 3, 4, and 5 mm) centripetally in the Eclipse™ Treatment Planning System.

2.C | Dosimetric effects of MRI distortion in liver SBRT plan
Five clinical liver SBRT cases were included in this retrospective study. The planning CT images of the liver SBRT plans were used to generate simulated "distorted" CT planning images. For "distortion simulation" method, the simulated "distorted" CT planning images were generated by applying the MRI distortion (as a deformation field) onto the planning CT images via deformable image registration.
For "body shrinkage" method, the simulated "distorted" CT planning images were generated by shrinking the body contour with a preset value (2, 3, 4, and 5 mm). For each liver SBRT case, we then generated two MRI-based liver SBRT plans using the simulated "distorted" CT planning images: Plan A, in which the dose is calculated on the simulated "dis-    In this study, we investigated the dosimetric effects caused by MRI geometric distortion in a computer simulation study. In addition, since the lack of electron density information in MRI is the other major impediments for MRI-based treatment planning, we also considered the dosimetric effects caused by CT number assignment.
We tried to use a more realistic geometric distortion, but due to

ACKNOWLEDG MENTS
This work is partly supported by a research grant from Varian.

CONFLI CT OF INTEREST
There is no conflict of interest involved in this study.