Program Information
Harmonic Analysis for Arbitrary 3D MR Image Distortion Fields
T Stanescu*, D Jaffray, Princess Margaret Cancer Centre, UHN, Toronto, ON; Radiation Oncology, University of Toronto, Toronto, ON; Techna Institute, UHN, Toronto, ON
Presentations
SU-F-J-154 (Sunday, July 31, 2016) 3:00 PM - 6:00 PM Room: Exhibit Hall
Purpose: To study the practical feasibility of harmonic analysis for the full quantification of 3D MR image distortion fields associated with arbitrarily-shaped regions of interest.
Methods: The 3D scanner-related distortion field (F) is innate to the MR image data. Since it is implicitly related to the magnetic fields required to perform MR imaging, F is in a steady-state and can be described by a boundary value problem in which the Laplace’s equation is solved with Dirichlet boundary conditions. Specifically, to fully derive a unique solution for F inside an arbitrarily-shaped domain (D) the minimum input data required is given by the boundary of D. In practice, the data that needs to be measured by means of image acquisition and post-processing is on the 3D surface of D (e.g. ellipsoid, cylinder, cuboid). Data from a large FOV phantom with a dense grid structure was used as a reference to explore multiple scenarios. The harmonic analysis solver was based on a finite element method and the domains were defined as binary STL structures for robust meshing. Simple and complex (non-)quadratic shapes were analyzed to test the applicability of the harmonic approach. The harmonic solution and reference data were compared inside D using histograms and typical metrics such as max/mean/stdev using a residual error threshold set by the accuracy of the reference data (i.e. 1 mm).
Results: The harmonic analysis showed good performance for all cases investigated. The percentage of control points inside D exhibiting residual values above 1 mm was lower than 1%.
Conclusion: Harmonic analysis can be used to characterize geometric distortions inside arbitrary volumes given availability of the surface data with a certain accuracy.
Contact Email: