next up previous contents
Next: Reformatting Up: Measurement in scanner Previous: Image acquisition

MR Geometrical Distortions

Geometrical distortion of MR images is rather significant in the context of . This section describes this phenomenon, along with some correction methods.

In MR imaging, linear field gradients allow each point in space to be related to a resonance frequency. In one dimension, say x, this can be expressed by:

where is the main magnetic field intensity, the gyromagnetic ratio of the imaged spin and the magnetic field gradient in the x direction. Since any conventional image reconstruction algorithm relies on this relationship, non-uniformity of , non-linearity of or susceptibility effects will introduce errors on the spatial localization of the different resonance frequencies.

It is possible to classify MR geometric distortion source in three main groups [41]:

Machine-dependent distortions factors, including main field inhomogeneity, gradient field non-linearity and eddy currents caused by the gradient switching, can in some cases be corrected for by careful calibration of the scanner and by phantom studies. It has been shown [6] that the contribution of gradient non-linearity to the distortion is stable and that it can be corrected for independently of the other distortion factors. Also, Michiels et al. [41] demonstrate that for 3-D spin-echo and gradient-echo sequences, the spatial misregistration introduced by main field inhomogeneity and gradient field non-linearity is:

where represents the main field inhomogeneity, the gradient field non-linearity and x is the direction of the read-out gradient. With the spin warp imaging method, each phase encoding step is performed with a different gradient magnitude (but the gradient is applied for the same time interval). So, the effect of field inhomogeneity in the phase encoding direction is to introduce a space-variant but time-invariant phase shift in that direction (since this gradient is turned off during readout). Since only the magnitude of the complex image is eventually displayed, this phase shift is of no consequence for the final image. Sumanaweera et al. [60] describe a method relying on this phenomenon to quantify field inhomogeneity. The principle of this technique is to acquire two images with the readout gradient in orthogonal directions. Any imaged object can be considered to be accurately positioned in the phase encoding direction. In this manner, the true position of the object can be found by combining the information given by the two images. Schad et al. [51] describes a geometric distortion study using phantoms of known geometry. One phantom was used to correct for in-plane distortion while another was used to correct for the slice profile distortion. This distortion correction reduced the error on the localization of a target using a stereotactic equipment from 3 mm to 1 mm, where the reference was the localization of the points using a CT scan. However, this method does not allow for correction of distortions induced by patient susceptibility variations. A comparison of the localization accuracy between CT and MR based stereotaxy, made by Kondziolka at al. [34], reports a localization difference around 2 mm between the two modalities. The authors note that the observed difference can be entirely attributed to observer variability in the selection of the target.

Patient-dependent factors, including susceptibility effects, chemical shift and flow, cannot be corrected ``once and for all'' since they depend on the particular anatomy of each patient. Distortion correction for objects of known shape and composition (fiducial markers for example) can be performed by means of an analysis of the susceptibility distortion induced by such objects. Sumanaweera [59] gives an example of this for a cylindrical fiducial marker. This technique requires the solution of the Laplace equation for the magnetic potential with boundary conditions, which implies that analytical solutions can only be obtained for highly symmetrical objects, and numerical methods must be employed for other cases. The application of this technique for highly variable shapes such as the human head, reported by Shizhe et al. [53], gives a good indication of the importance of local magnetic field distortion that can be induced by the patient.



next up previous contents
Next: Reformatting Up: Measurement in scanner Previous: Image acquisition



Patrice MUNGER
Mon Oct 23 15:09:17 EDT 1995