Magnetic resonance imaging (MRI) based radiation treatment planning has gained wide clinical acceptance due to the superior soft tissue contrast when compared to x-ray computed tomography (CT). MRI is particularly attractive for proton beam therapy because of the need to clearly demarcate the target from surrounding radiosensitive organs at risk. There is also interest in using MR information only for both photon and proton radiation therapy planning [1], as well as for attenuation correction for PET/MR systems[2,3]. The main technical challenge in utilizing this information lies in determining the accurate and precise geometric precision of the tissues that are to be targeted or avoided during radiation therapy.

A potential source of error that can affect the geometric precision of an MR image is spatial distortion induced by spatial gradient nonlinearity (GNL). Imaging gradients are designed to vary linearly across the field-of-view (FOV), but the Maxwell equations and engineering limitations dictate that there will be some residual GNL, usually manifesting as a “pincushion” type of effect. This effect is more pronounced at the edges of the FOV when a large FOV is prescribed, with position errors of up to 5 mm from truth potentially impairing the accuracy of treatment planning. Previous work targeting a small FOV imaging device utilized up to 10th order GNL correction with spherical harmonic coefficients obtained via an iterative calibration procedure and use of the ADNI [4]. This work extends that idea to whole body systems with large FOV imaging.

This work utilizes a large fiducial phantom [5] as shown in Figure 1. The phantom consists of a matrix of MRI visible paintballs uniformly distributed throughout. Each paintball is separated by 2.5 cm from its neighbor in each direction. Images of the phantom were acquired with a fast gradient-echo sequence with 61.4 cm square FOV, 1.2 mm isotropic voxels and ±125 kHz receiver bandwidth utilizing the body coil for transmit and receive on a 3T scanner (GE 750w, DV25.0, GE Healthcare, Waukesha WI) with a 50cm imaging diameter spherical volume (DSV) and 70cm patient bore aperture.

A 3D Hough-transform based position tracking software was developed in house to match both the MRI data with a CT reference scan to map GNL-induced spatial distortions. These positions were compared, and used to generate high order (up to 9th) corrections utilizing the iterative gradient nonlinearity estimation framework [4]. We utilized this framework to generate 3rd, 5th, 7th, and 9th order coefficients, and then applied these coefficients to reconstruct a series of corrected images. Finally, the error of each fiducial marker interrogated by the fitting routine was plotted to generate position dependent accuracy figures.

Figure 2 shows spatial accuracy in the form of RMSE vs. the CT position reference for a 280mm diameter spherical volume (DSV), comparing the fitted 5th, 7th, and 9th order corrections. Additionally, Figure 3 shows the fitted 5,7,9th order corrections for a 500mm DSV.

Figure 4 shows axial and coronal planes at approximately 14 cm inferior to isocenter. Note the better depiction of planar “paintballs” in the 7th and 9th order case, vs 5th order.

Figure 5 shows the RMSE comparison as a function of spherical harmonic model order used. Note the 9th order showing minimum RMSE.

We have demonstrated an imaging-based method for characterizing the system specific gradient non-linearity of whole-body scanners with large FOV acquisition capabilities. This reduces the RMSE error substantially from standard 5th order GNL correction, potentially allowing for a higher level of confidence in utilizing MR images for radiation therapy planning. Note in Figure 4 the increased background noise to the identically window/leveled images with the higher order correction. This artifact is due to an increase compensation term from “spreading out” image intensity in the image-based correction for gradient nonlinearity. An integrated gradient nonlinearity correction and image reconstruction [6] is expected to further reduce this effect.

The outcome of this work is an expected increase in geometric accuracy and precision confidence in proton beam therapy planning. In the future, multiple systems will be calibrated using the described procedure to account for system-specific deviations from the standard electromagnetic field simulated coefficients in order to examine the significance of system-by-system deviation.