Biomedical Engineering Reference
In-Depth Information
What if we are using image registration to monitor change? If the change is
only in intensity and we can guarantee that the imaged structure has not
changed size, as might be the case in functional neuroimaging of the brain by
MRI (fMRI), then the concept of correspondence is straightforward. If, how-
ever, change in volume is possible, such as when monitoring atrophic
changes in the brain in longterm studies, then the concept of correspondence
becomes less well defined. We are using serial MR to monitor neuronal loss,
so the tissue corresponding to a certain location in the patient at the first visit
may not exist at the later visit. In this case we normally undertake registra-
tion and establish correspondence, assuming that the tissue has not changed
volume, and change will be inferred from absence of correspondence, often
by displacements of boundaries. This may be displayed to good effect by
using image subtraction, as explained in more detail in Chapter 7. Detecting
a change in boundary location is relatively straightforward. Inferring which
particular element of tissue has changed volume is much more difficult. Even
in our first example, workup for neurosurgery, changes may occur due to dif-
ferences in patient positioning, systemic blood pressure, hydration, and
imaging at different phases of the breathing or cardiac cycle, although most
of these changes are thought to be below the level detectable with current
imaging devices.
When tissue changes significantly over time, such as long term monitoring
of the breast postmenopause with MR, the concept of correspondence becomes
even more difficult to define. Over repeat scans for a year or two, glandular
tissue may be replaced by fatty tissue, yet registration to establish anatomical
correspondence might have great potential to detect the presence of small
focal disease, as described in Chapter 13. In this case the required transforma-
tion has to allow the tissue to deform by changing shape and size. This type
of transformation is called “nonrigid” and is more appropriate for soft tis-
sues. Over a short time period, for example, a dynamic acquisition of 3D MR
images taken over a few tens of minutes, tissue will obey the laws of physical
motion associated with elasticity and viscosity; during longer periods of
time, tissue may grow or be destroyed. When monitoring surgery or thera-
peutic response, tissue will, by definition, be removed, and therefore corre-
spondence in the excised or ablated region cannot exist. Conversely, the
growth of a tumor adds tissue and structure. Anesthetics and drugs to control
edema may also change tissue volumes.
Finally, we need to consider correspondence when combining images
from multiple subjects or images between an individual and an atlas
derived from one or more other individuals. Is correspondence defined in
anatomical terms, e.g., the most anterior pole of the occipital cerebral cor-
tex? Or is it defined in geometric terms, e.g., the portion of cerebral cortex
in this region with the highest surface curvature? Is it defined in functional
terms, e.g., the region of the cerebral cortex associated with vision that
exhibits the greatest change in blood flow when the subject observes a flash-
ing checkerboard pattern? Or is it defined by a particular histological
appearance? On a coarse scale these may all correspond in normal subjects
