Biomedical Engineering Reference
In-Depth Information
27
was used by Jezzard and Balaban to correct (“unwarp”) EPI data.
Reber et al.
use field maps obtained using high signal-to-noise ratio (SNR) EPI data for
increased robustness.
30
Kadah and Hu propose an algebraic approach to the
problem of identifying the optimal inhomogeneity distortion operator.
31
Ernst et al. propose coregistration with a reference (undistorted, non-EPI)
image dataset as a correction method for EPI distortion.
32
The method is
based on matching the brain surface to correct for global scaling and shearing
errors in the EPI data. However, it cannot correct for distortion and signal loss
due to local field inhomogeneity.
5.3.4
Error in Field of View (Voxel Dimensions) Due to Variations
in Gradient Strength
Important fluctuations in imaging gradient amplitudes from prescribed values
are common.
33
Furthermore, system calibration tolerances are typically much
larger than stability effects. In frame-based stereotaxy, such fluctuations are
largely irrelevant since the scaling factors can be derived from the frame. In the
absence of a frame, the impact of uncertainty in the voxel dimensions is poten-
tially important in certain applications of image registration, in particular lon-
gitudinal studies of morphological changes in the brain, image-guided
neurosurgery combining CT and MRI, and radiotherapy planning.
5.3.4.1
Correction
Registration methods that incorporate linear scaling factors can compensate
for such differences assuming that the dimensions of features used for match-
ing have not varied in the interscan interval. To this effect the skull can be
used as such an invariant structure and Freeborough and Fox have been able
to observe very small changes in brain volumes by this means.
34
Alternatively
a method based on the registration of a test object as part of a QA schedule
has been proposed.
35
In particular, it has been observed that recalibration
during normal servicing can introduce important variations in the scaling
factors. The derived scaling factors can then be used as fixed factors in a rigid-
body registration of head scans. Hill et al. proposed a method to measure the
actual errors in the voxel dimensions based on scanning a purpose-built test
object and registering the resulting images with a computerized version of
the phantom.
The method was devised for application in image-guided neu-
rosurgery using MRI and CT. A significant improvement in registration accu-
racy was obtained when using the scaling factors derived from the phantom
when compared with the “free,” nine-degrees-of-freedom registration.
36
5.3.5
Signal Nonuniformity Due to RF Inhomogeneity
Signal nonuniformity can be present in all MR images and in particular
those acquired with smaller RF coils, such as surface coils.
37
Although RF
inhomogeneity does not give rise to geometric distortions, the ensuing
signal nonuniformity can be a significant problem for image registration.
