Biomedical Engineering Reference
In-Depth Information
venous cannula and the rate of blood flow. The best results are usually obtained
from a cannula that is placed in a large vein on the dorsum of the hand, typically
in a muscular male. As mentioned earlier in this section, tracer concentration
in venous blood may differ markedly from that in arterial blood, especially for
tracer with high extraction fractions, and the use of venous blood as the input
function could introduce significant errors in absolute quantification of physio-
logical parameters.
When a suitable vascular structure, such as a large artery or the left ventricle,
is contained within the FOV, the input functions can be derived noninvasively
from the image data and the need for blood sampling is completely eliminated.
However, careful correction for the extravascular contamination in the blood
region is required to obtain an accurate approximation to the input function.
In certain circumstances, if regions can be identified with the same input but
with different kinetic behavior (e.g. gray and white matter in the brain), then
the compartment model can be reformulated to solve for the kinetic parameters
of interest and the required input function [69-71]. However, these approaches
may have very high computational complexity as the number of parameters to
be estimated (kinetic parameters and the input function) increased drastically
and careful selection of parameter estimation algorithm may be required.
In the case of receptor studies, the need for measuring arterial input function
may be avoided if a tissue region can be identified which is devoid of specific
binding but shows similar nonspecific binding to the tissues of interest. Typically,
the cerebellum is used as a reference tissue region in the brain for a number
of neuroreceptor systems. Once the reference tissue region is identified, the
compartmental model can be reformulated to use the reference tissue TAC as
the input function [72-74]. The use of reference tissue also eliminates the need
to estimate the fraction of labeled metabolites present in the plasma.
2.14.4 Parametric Imaging
PET (and SPECT) provide multidimensional images of the time-varying radio-
pharmaceutical distribution. The aforementioned compartmental analysis has
concentrated only on fitting tissue TACs derived from ROIs to a predefined com-
partment model. Although it is possible to generate TACs for each voxel of the
image data and fit them to the model, this is not practical as many thousands
of model fit using NLLS, which is computationally expensive, would have to
be performed. Further, the high noise levels and heterogeneity in individual
