Compensating Small Head Displacements for an accurate fMRI Registration
Abstract
Many medical image processing algorithms require the comparison of multiple scans acquired from a given subject. In the absence of an absolute coordinate system, these different images have to be aligned in a common space before being able to identify similar anatomical landmarks. Several rigid registration techniques have been proposed that estimate the rigid transformation between two anatomical volumes [1] but none of them introduce geometric constraints on the possible position of the human head inside a scanner. We propose a new method to improve the accuracy of rigid registration in the case of small displacements by restricting the number of degrees of freedom (dof) involved in the registration process. It has been shown in a different context that decreasing the number of dof improves the numerical stability of the displacement recovery algorithm [16, 8] by regularizing the minimized criterion and reducing the number of local minima. We developed a strategy adapted to the case of human head displacement in a MRI scanner. This algorithm has been successfully tested on 6 sets of fMRI volumes acquired at the MNI on different patients. Each set is composed of 30 to 78 volumes.
Origin : Files produced by the author(s)
Loading...