Evaluation of cine imaging during multileaf collimator and gantry motion for real‐time magnetic resonance guided radiation therapy

Abstract Purpose Real‐time magnetic resonance guided radiation therapy (MRgRT) uses 2D cine imaging for target tracking. This work evaluates the percent image uniformity (PIU) and spatial integrity of cine images in the presence of multileaf collimator (MLC) and gantry motion in order to simulate sliding window and volumetric modulated arc therapy (VMAT) conditions. Methods Percent image uniformity and spatial integrity of cine images were measured (1) during MLC motion, (2) as a function of static gantry position, and (3) during gantry rotation. PIU was calculated according to the ACR MRI Quality Control Manual. Spatial integrity was evaluated by measuring the geometric distortion of 16 measured marker positions (10 cm or 15.225 cm from isocenter). Results The PIU of cine images did not vary by more than 1% from static linac conditions during MLC motion and did not vary by more than 3% during gantry rotation. Banding artifacts were present during gantry rotation. The geometric distortion in the cine images was less than 0.88 mm for all points measured throughout MLC motion. For all static gantry positions, the geometric distortion was less than 0.88 mm at 10 cm from isocenter and less than 1.4 mm at 15.225 cm from isocenter. During gantry rotation, the geometric distortion remained less than 0.92 mm at 10 cm from isocenter and less than 1.60 mm at 15.225 cm from isocenter. Conclusion During MLC motion, cine images maintained adequate PIU, and the geometric distortion of points within 15.225 cm from isocenter was less than the 1 mm threshold necessary for real‐time target tracking and gating. During gantry rotation, PIU was negatively affected by banding artifacts, and spatial integrity was only maintained within 10 cm from isocenter. Future work should investigate the effects imaging artifacts have on real‐time target tracking during MRgRT.

The ViewRay MRIdian (ViewRay Inc, Oakwood Village, OH, USA) system integrates a magnetic resonance imaging (MRI) unit with a linear accelerator (linac) to deliver MR-guided radiation therapy (MRgRT).
The use of MRgRT allows tumors to be visualized in real time during treatment with excellent soft tissue differentiation and no additional exposure to ionizing radiation. Gating of the beam based on tracking of internal anatomy during treatment spares normal tissue and ensures the tumor is not underdosed due to patient motion. Without MRgRT, an additional margin is added to the clinical target volume to account for internal motion during treatment. This internal target volume (ITV) ensures that the tumor reliably receives the appropriate dose, but also results in greater normal tissue irradiation.
There are currently several technological challenges that result from the integration of a linac with an MRI system. 1  inhomogeneities, and nonlinear gradients. Each of these can result in geometric distortions that affect the spatial integrity of the image. 2,3 Additionally, eddy currents produce banding artifacts that interfere with the signal intensity. Deviations in signal intensity that are not related to patient anatomy can result in target tracking errors. 1,[4][5][6] For these reasons, real-time MRgRT systems use step-and-shoot IMRT, in which the gantry and MLC are static during delivery, to maintain image quality during treatment. As a consequence, the advantages that sliding window IMRT and volumetric modulated arc therapy (VMAT) have over step-and-shoot delivery, specifically plan quality and treatment time, are not currently available in MRgRT.
Both sliding window IMRT and VMAT allow more freedom to modulate beams for conformal shaping of the dose distribution than is available in step-and-shoot IMRT. 7 Each has been shown to produce treatment plans that are dosimetrically superior to those of step-and-shoot IMRT, although the results are more modest for sliding window. 8,9 The dosimetric improvements include better dose homogeneity, PTV coverage, and OAR sparing. This has, in part, been used as an argument for VMAT being a better choice than MRgRT for certain treatments such as lung SBRT. 10 A second advantage sliding window IMRT and VMAT have over step-and-shoot delivery is reduced treatment time. 8 Faster treatments have the benefits of increasing throughput (so that more patients can benefit from the treatment) and decreasing discomfort (from, e.g., positional strain or a full bladder).

2.A | MRIdian system overview
The ViewRay MRIdian system combines a 0.35 T split bore superconducting magnet, a 6 MV flattening filter free (FFF) standing wave linear accelerator, and a fully integrated adaptive treatment planning system. The system uses a 70 cm bore and has 20 cm-50 cm diameter spherical field of view (FOV). The radiation therapy system can deliver a dose rate of 650 cGy/min at the source-to-axis distance of 90 cm. A double-stack, double-focus MLC is used to shape the beam for either 3D conformal or step-and-shoot IMRT delivery. The treatment planning system calculates dose using a Monte Carlo algorithm that accounts for effects of the 0.35 T magnetic field.   Based on characterization of our system using a ViewRay provided technical service report, we found that the minimum isocenter shift resulting from varying gantry positions occurs when the gantry is at 330°, the 3D static gantry data collected at 330°were used as the true phantom position when determining spatial deviation. As with the PIU measurement, the moving gantry data were collected by imaging in cine mode while the gantry was moved from 360°to 90°, and the moving MLC data were collected by imaging in cine mode while running the MLC positional second check. The uniform linearity phantom was aligned in the sagittal plane and centered with the virtual isocenter in accordance with the spatial integrity test done during monthly quality assurance (QA). 3

2.D | Percent image uniformity analysis
Image uniformity was evaluated by calculating the percent image uniformity.
where MaxROI and MinROI were found using the procedure outlined in the American College of Radiology MRI Quality Control Manual. 15 Briefly, for both 3D and cine images, a 2D mean signal ROI that was concentric with the phantom and covered 75% of the phantom's cross-sectional area was created. The measurements to determine the MaxROI and the MinROI were taken within this area. The Max-ROI was found by adjusting the window level so that only a small number of bright pixels were visible within the phantom (Fig. 2a). A small measurement ROI with an area that is 0.15% of the FOV area was generated to include the brightest pixels. The mean signal value within the measurement ROI was the MaxROI. The MinROI was found in the same way, except that the window level was adjusted so that only a small number of the darkest pixels were visible within the phantom (Fig 2b). The ACR recommends a PIU of 87.5% and establishes 85% as passing. 15 However, these tolerances are established for ACR recommended sequences used for monthly QA rather than clinical scans. Therefore, the PIU of the cine images acquired during linac component motion were compared to the PIU of the cine images acquired under the static linac conditions currently used for treatment. ViewRay provides a tool that automates this for 3D images (it is not designed to be used on 2D cine images). It is possible to automate marker localization for the 2D cine images as well, however the banding artifacts that occur during linac component motion make it particularly challenging to consistently identify makers and their positions. Therefore, the analysis of the 2D cine spatial distortion was done in accordance with the imaging characterization performed by Hu et al 5 Measurement locations were chosen to lie on two circles centered at the isocenter with radii of 10 cm and 15.225 cm.

2.E | Spatial integrity analysis
Eight measurement locations spaced every 45°were located on each circle (Fig. 3a). The larger radius (15.225 cm) corresponds to the outermost rows and columns of markers in the phantom, making it the largest circle that will allow 8 measurement points. If a measurement location was between markers, the nearest marker was used.
Hu, et al. used a larger spatial integrity phantom which allowed them to use a radius of 17.5 cm for the outer circle (note that this distance from isocenter corresponds to a change in the passing threshold from 1 mm deviation to 2 mm deviation for monthly QA). During MRgRT, the most critical tracking structures lie well within a 15 cm radius, and therefore the data collected in this study are representative of clinical situations.
In order to improve marker localization, 2D cine images were upsampled using the bicubic method from the native planar resolu-

3.A | Percent image uniformity
Prior to the moving MLC test, the 3D static PIU was measured to be 88.3%, and the 2D cine static PIU was measured to be 85.6%. 4 The variation of the PIU of the 2D cine images from the static value as the MLC leaves move is presented in Fig. 4a  isocenter) measurement points. The mean deviation is less than 0.45 mm for the inner points throughout the MLC motion, and the maximum deviation for those points was 0.72 mm (Fig. 5b). The mean deviation for the outer points did not exceed 0.60 mm through the MLC motion, and the maximum deviation for those points was 0.88 mm (Fig. 5c). Fig. 6 shows the displacement of the imaging isocenter, which was measured as the translational displacement from the 3D high resolution image collected at 330°. Throughout MLC motion, the isocenter deviation did not exceed 0.12 mm in both the longitudinal (Y) and vertical (Z) directions.

3.B.2 | Static gantry
The mean geometric distortions as a function of static gantry position for both the inner and outer measurement points are plotted in Fig. 7. The mean geometric distortion for the inner points was greatest at a gantry angle of 90°. At this position, the mean and maximum distortions were 0.65 mm and 0.88 mm (Fig. 7b), respectively. The greatest distortion for the outer points occurred at 240°, where the mean and maximum were 0.92 mm and 1.4 mm (Fig. 7c), respec- tively. The geometric distortion shows a gantry-position-dependent behavior with minima at 150°and 330°, a maximum at 240°, and apparent approach toward another maximum between 360°and 90°.
The shift in the imaging isocenter is shown in Fig. 8.

3.B.3 | Moving gantry
As with the static gantry data, the mean geometric distortion as a function of moving gantry position is displayed in Fig. 9. The mean geometric distortion was greatest at a gantry angle of 240°. At this position, the mean and maximum distortions for the inner points were 0.77 mm and 0.92 mm (Fig 9b) respectively. The mean and maximum distortions for outer points were 0.95 mm and 1.6 mm ( Fig. 9c), respectively. Again, the geometric distortion shows a gantry-position-dependent behavior. The minima are slightly displaced (135°and 345°, cf 150°and 330°). The local maximum at 240°is still present, though, as is the apparent approach toward another maximum between 360°and 90°.
The shift in the imaging isocenter is shown in Fig. 10.  The results of our image quality tests performed during MLC motion show that the PIU does not vary by more than AE1% from the static linac conditions currently used during treatment (Fig. 4a).
No banding artifacts were observed that would jeopardize target tracking during treatment. The mean geometric distortion of points 10 cm from isocenter was less than 0.4 mm for every image analyzed throughout the MLC motion (Fig. 5), and the maximum deviation was 0.72 mm. The mean geometric distortion for points 15.225 cm from isocenter was less than 0.60 mm (Fig. 5), and the maximum deviation was 0.88 mm. Therefore, all geometric distortions measured within 15.225 cm from isocenter do not exceed the 1 mm threshold for acceptable spatial integrity.
The image quality tests performed during gantry rotation from 360°to 90°show that the PIU variation relative to static conditions ranged from −1.6% to 2.7% (Fig. 4b), and this increased variation was due in part to banding artifacts present during the movement of the gantry (Fig. 11). The signal intensity variations resulting from these artifacts could prevent reliable tracking of target structures during MRgRT.
Characterization of the effect of gantry motion on spatial integrity was more involved than in the case of moving MLC because it is not only the gantry motion that can affect image quality, but the static gantry position as well. The gantry does not have a uniform magnetic susceptibility, and therefore its static position may affect the magnetic field homogeneity, which in turn affects geometric distortion. 2 Therefore geometric distortion analysis was done for both the observed maximum at 240°and 90°from each of the observed minima at 150°and 330°. This should, however, be verified. For the measurement points lying 10 cm from isocenter, the mean distortion was less than 0.65 mm for all gantry positions (Fig. 7), and the maximum deviation was 0.88 mm. Therefore, at 10 cm from isocenter, all deviations measured on the 2D cine image were within the 1 mm limit considered acceptable for MRgRT. For the measurement points located 15.225 cm from isocenter, the mean distortion was less than 0.92 mm for all gantry positions (Fig. 7), and the maximum deviation displacement and geometric distortion, the static gantry data were subtracted from the moving gantry data for each (Fig. 12). In The image distortions discussed in this paper only represent those that result from the ViewRay system. During cine imaging and target tracking, additional patient-induced distortions also arise. 3,16,17 These are caused by chemical shifts, variations in magnetic susceptibility, and patient movement. The effects these distortions have on cine tracking will need to be investigated, and those results would need to be incorporated into any evaluation of the accuracy of cine target tracking.  (2)  It is worth noting that there are patient-specific sources of image distortion, namely chemical shifts and magnetic susceptibility, that cannot be evaluated with phantom studies. 2 We were determined to collect data for ≈ 270 s of continuous acquisition. This corresponded to ≈ 270°of gantry rotation. Since ViewRay imaging is subject to a gantry limit switch, we chose 360°-90°as the range of gantry rotation for these measurements. 3 While our monthly quality assurance (QA) procedure evaluates spatial integrity in all three anatomical planes, only the sagittal plane is relevant to 2D cine imaging. 4 Note that the ROIs used to calculate PIU are taken within a volume of uniform magnetic susceptibility. Therefore, a higher contrast, but less noisy, 3D scan sequence may have a higher PIU than a lower contrast, but noisier, 2D cine sequence.R E F E R -E N C E S