< Back to topics
> Cover Page, Authors, Institutions
> View tables/images/videos
on a method proposed by Langlois et al. , a cubic, oil-filled phantom
fitting inside the head coil of the scanner (Siemens Maestro Sonata) is scanned
with three 3D mpr sequences (axial, coronal, and sagittal) with a voxel size of
0.4 × 0.4 × 2 mm3. The edges of the phantom are detected using an
advanced edge detection algorithm (Matlab) and unwanted image features are
either removed (e.g., partial phantom images) or repaired (e.g., air bubbles).
For each pair of phantom edges midpoints are calculated, which are then used to
reconstruct undistorted midplanes of the phantom. Shifting the midplanes by
half the known phantom diameter along their normal vectors in each direction
yields estimates for the undistorted phantom planes. Distortion correction
functions are modeled as sums of spherical harmonics up to the 5th order.
Function parameters are then obtained by a least-squares fitting algorithm.
Knowing the distortion correction functions, the gradient nonlinearties of any
MRI study produced by the scanner can be removed.