45 Quantification of geometric distortion on MR images and evaluation of the impact of distortion correction

Introduction Magnetic resonance (MR) images have proven very useful for target definition in radiotherapy. However, geometric distortions might be of concern. In this study, a GE software based solution to correct geometric distortions is evaluated. The amount of residual distortions on corrected MRI is then quantified and used to determine PTV margins. Methods Axial T1 gradient-echo sequences of an MR compatible phantom and of patients with brain metastases were acquired on a 3T MRI. All patients were injected with Gadolinium. MR images were corrected for geometric distortion using the default algorithm implemented on our acquisition software, leading to two sets of MR images per acquisition (corrected and uncorrected). The gross tumor volume (GTV) was then contoured by a physician on both corrected and uncorrected MR images of the patients. Acquisition parameters were the ones commonly used in clinical practice for intracranial tumors. The MR phantom external contour was drawn on both corrected and uncorrected MR images, and used to deform uncorrected images to match corrected images. The amplitude of the displacement vector field was used to identify the main regions of the image undergoing large geometric distortions (Figs. 1 and 2). The same work was done on patients using the external surface of the patients. Geometric distortions were quantified using visible markers on both CT and MR images (Fig. 3). Results The main regions affected by geometric distortion were identified using the vector displacement field. These regions are located apart from the machine isocenter in the transverse direction (Figs. 4 and 5). Maximum distortions measured on the phantom for uncorrected MR images ranged from 1 to 2 mm for slices located near the isocenter and until 3 mm for the most distant slices. Distortions remained below 1 mm in corrected MRI. Considering patient images, a maximum distance of 2 mm was observed between the GTVs contoured on both corrected and uncorrected MRI (Fig. 6). Conclusions Our results point out the importance of using a correction for geometric distortions in MR images. Without being corrected for, distortions might reach several millimeters. The correction algorithm we used significantly reduces the amount of distortion in the images, particularly in regions located away from the isocenter. Quantification of residual distortions was made on phantom. However, additional sources of distortions might be of interest when patients are considered as for example the difference in magnetic susceptibility among tissues. We are currently investigating a multimodal approach based on CT-MRI registration to quantify patient-specific distortions. Download high-res image (570KB) Download full-size image