SU-E-J-227: Evaluation of Residual Geometric Distortion in MRI for Treatment Planning

Purpose: To determine an appropriate model to simulate system related geometric distortions in MR images in order to investigate the dosimetric impact on MRI-LINAC treatment planning. Methods: Two sets of geometric distortion data in 3D volume were obtained independently from two different MRI vendors using different geometric distortion phantoms with visible markers. The distortion data were used to train two parametric models in X, Y, and Z directions separately: a second order polynomial model and a spherical harmonic model. The polynomial model was constructed in 2D slices with available measurements while the spherical harmonic model was built in 3D. After the models were created, they were used to generate simulated distortions at the marker positions to be compared with the measured data for evaluation. Results: The maximum measured distortions (after standard distortion corrections) were at the border of the scan, with 6.9 mm, 6.1 mm, and 3.6 mm in the X, Y, Z (slice) direction for first dataset, and 21.7 mm, 4.0 mm, and 12.0 mm for second dataset, respectively. For the first dataset, the polynomial model had average simulation errors in X, Y, and Z of 0.4±0.4 mm, 0.2±0.2 mm, and 0.1±0.2 mm, contrasting with those from the spherical harmonic model of 0.5±0.7 mm, 0.4±0.4 mm, and 0.6±0.4 mm. For the second dataset, the polynomial model had average simulation errors of 1.4±1.8 mm, 0.3±0.3 mm, and 0.4±0.5 mm, contrasting with those from the spherical harmonic model of 2.8±3.2 mm, 0.6±0.5 mm, and 2.5±2.1 mm. We found that using a maximum order of 2 in the spherical harmonic model gave the best estimation of the distortion. Conclusion: Polynomial model showed a better simulation of the residual geometric distortion than spherical harmonic model and may be a better choice to model the residual MR image distortion for MRI-LINAC treatment planning purposes. This work was partially supported by a research grant from Elekta.