Biomedical Engineering Reference
In-Depth Information
FIGURE 4.3
When data is transferred from one computer to another of a different make it is often
necessary to reverse the byte order of the numbers stored for each pixel value. (a) An image
of the brain stored and displayed on a Sun workstation as an array of short integers with
two bytes per pixel, (b) the same image copied to a Compaq Alpha workstation, and (c) the
image in (b) after byte reversal again displayed in the Compaq computer.
in different orientations depending on the current settings. On-screen labels
are generally provided to remove any ambiguity. However, image registra-
tion software may be required to align images originally stored with different
coordinate systems, so care is required to ensure consistency. Use of images
from test objects in which left-right, head-foot, and anterior-posterior orien-
tations can be unambiguously determined provides a simple robust method
of ensuring that all data are presented to the registration algorithms and sub-
sequent display system with appropriate consistency. An advantage of a
comprehensive format such as DICOM3 is that it retains image orientation
information that is likely to be lost in simpler formats.
4.4
Intensity, Size, and Skew Correction
Having achieved a consistent data format and a standardized convention
for the orientation of the data, it may still be necessary to manipulate the
images prior to image registration. Errors in scanner calibration or image
artifacts may need to be dealt with (see Chapter 5). Even with ideal perfor-
mance, the acquisition techniques may have implications for subsequent
processing. Examples include changes in global intensity scaling during
segmented acquisition with radiotracers, or variation of intensity across
images produced by the use of surface coils in MRI. Algorithms designed to
register similar images such as those based on minimization of least-square-
intensity differences
6,7
may not work robustly if the images have differently
scaled intensity maps. This is easily corrected by manually or automatically
rescaling the images to matched intensity ranges. Most current registration
algorithms are not robust in the presence of large intensity variations across
the images, so it may be necessary to apply intensity correction schemes prior
to registration. MR scanners as well as post processing packages frequently
