Analysis of machine-dependent and object-induced geometric distortion in 2DFT MR imaging

Geometric distortion in MR imaging predominantly arises from the inhomogeneity of the static field and the nonlinearity of the gradients. It is the purpose of this paper to analyse the object and machine related contributions to geometric distortion in order to determine which corrections are necessary for attaining a specified precision. System related imperfections were measured by systematic variation of the strength, direction, and polarity of the read-out gradient in imaging experiments on a grid of cylindrical sample tubes. For the 1.5-T system used in this study, static field related errors up to 7 mm and gradient related errors up to 4 mm were observed (midcoronal plane, FOV 400-mm, G-read between 0.5 and 3.0 mT/m). Field related errors were shown to be inversely proportional to gradient strength, whereas gradient related errors turned out to be virtually independent of gradient strength. It therefore seems recommendable to always apply the strongest available selection and read-out gradients when geometric fidelity is given preference to signal-to-noise considerations. Correction of system related geometric distortions in MR images can readily be performed by table lookup. Object-induced distortions of the gradient fields were studied by experiments on a grid of sample tubes immersed into a cylindrical water bath of variable saline concentration. These experiments revealed a negligible influence of the object on the gradient error distribution, and lead to the conclusion that correction for the nonlinearity of the gradients only requires the application of system dependent correction factors. Object-related distortions of B0 were studied by conventional SE and fat-suppressed IR experiments on phantoms and human subjects. In these experiments the polarity of the read-out gradient was reversed. Subtraction images showed significant object-induced inhomogeneities of the static field at tissue-air interfaces and in the immediate vicinity of the object being imaged. A first attempt to correct for object related B0 inhomogeneities was made by contour analysis of the source images. At present this correction still has to be done manually.