MRI simulation: effect of gradient distortions on three-dimensional prostate cancer plans.

PURPOSE To quantify the dosimetric consequences of external patient contour distortions produced on low-field and high-field MRIs for external beam radiation of prostate cancer. METHODS AND MATERIALS A linearity phantom consisting of a grid filled with contrast material was scanned on a spiral CT, a 0.23 T open MRI, and a 1.5 T closed bore system. Subsequently, 12 patients with prostate cancer were scanned on CT and the open MRI. A gradient distortion correction (GDC) program was used to postprocess the MRI images. Eight of the patients were also scanned on the 1.5 T MRI with integrated GDC correction. All data sets were fused according to their bony landmarks using a chamfer-matching algorithm. The prostate volume was contoured on an MRI image, irrespective of the apparent prostate location in those sets. Thus, the same target volume was planned and used for calculating the anterior-posterior (AP) and lateral separations. The number of monitor units required for treatment using a four-field conformal technique was compared. Because there are also setup variations in patient outer contours, two different CT scans from 20 different patients were fused, and the differences in AP and lateral separations were measured to obtain an estimate of the mean interfractional separation variation. RESULTS All AP separations measured on MRI were statistically indistinguishable from those on CT within the interfractional separation variations. The mean differences between CT and low-field MRI and CT and high-field MRI lateral separations were 1.6 cm and 0.7 cm, respectively, and were statistically significantly different from zero. However, after the GDC was applied to the low-field images, the difference became 0.4 +/- 0.4 mm (mean +/- standard deviation), which was statistically insignificant from the CT-to-CT variations. The mean variations in the lateral separations from the low-field images with GDC would result in a dosimetric difference of <1%, assuming an equally weighted four-field 18-MV technique for patient separations up to approximately 40 cm. CONCLUSIONS For patients with lateral separations <40 cm, a homogeneous calculation simulated using a 1.5 T MRI or a 0.23 T MRI with a gradient distortion correction will yield a monitor unit calculation indistinguishable from that generated using CT simulation.