A Novel Automated Approach for the Evaluation and Correction of Geometric Distortion in Magnetic Resonance Imaging

Background: The geometric distortion of magnetic resonance (MR) images, mainly due to the inhomogeneity of the magnetic field, is a limiting factor for MR-only radiotherapy. Aims: This study presents a new automated method to identify and correct MR imaging (MRI) image distortion. Methods: A large 3D home-made phantom with cylindrical water pipes has been used for the geometric distortion measurements of three MR protocols: turbo inversion recovery magnitude (TIRM), turbo spin-echo (TSE), and fast low angle shot (FLASH) sequences. The vascular modeling toolkit, 3D Slicer, and MATLAB software have been used to determine the correction matrix of MR geometric distortion. Results: The maximum mean absolute error (MAE) of geometric distortion has been achieved 8.17, 4.25, and 7.12 in TIRM, TSE, and FLASH MR sequences, respectively, that has been reduced to 2.10, 0.57, and 0.38 after the automated correction of geometric distortion. In TIRM and TSE protocols, the phase-encoding direction showed more distortion than the frequency-encoding direction. The maximum and minimum MAE values of 8.17 and 1.72 mm has been seen in the phase- and frequency-encoding direction of TIRM and TSE sequences, respectively. Conclusion: The observed distortion of MR images depends on the image protocol and the slice position, resulting in a unique conversion matrix for each MR image sequence. The automated method proposed in this research can effectively assess and rectify geometric distortions in MR images for large field of views of MRI-only radiotherapy purposes.