A new approach to estimation of non-isotropic scale factors for correction of MR distortion

Purpose: When performing image-guided neurosurgery, MR images are widely applied for the planning of surgical path. However, a MR image sometimes suffers geometry distortion, limiting the surgical outcome. Correction of geometry distortions are thus performed prior to the surgical operation, which is normally in the reference of CT images. Usually distortions can be system inherent, e.g., field inhomogeneity, or patient induced, such as wearing implantable devices, and are detected using the fiducial markers from a head frame. By registration of the markers located from both MR and CT images, it is expected the distorted or transformed parameters from MR images can be found. As such, most existing approaches apply the work developed by Arun et al to locate translate and rotate matrixes using least-squares technique, which however does not take scale transformation into account and has since been extended to include an isotropic scaling. In our study, it is found that the scale factors are not the same along 3 axial directions of a MR image, i.e, with nonisotropic scale, necessitating the need to find scale matrix as well as the other transformation matrixes.