Biomedical Engineering Reference
In-Depth Information
Use of biophysical models
It is partly because of the arbitrary nature of the dose-based measures
identified above, and their almost unknowable relative weighting
factors,
3
that interest has grown in the use of biological models such
as TCP, NTCP, and EUD for optimization. While confidence in the
estimation of these quantities may be poor, this is compensated by the
fact that
they have an intuitively obvious clinical meaning
. As a
result, the relative importance of a given increase in one of them, say
the TCP, and in another, say the NTCP for pneumonitis, can be
readily understood in human terms. Indeed, these are quantities that
the patient too can understand and the relative weighting of which he
or she may have an opinion about which should be taken into account.
Use of a score in optimization
There are two ways to use a score in the process of optimization:
Score optimization
In this approach all the variables
to be determined are bundled into the score function and the
optimizer must choose the values of those variables that maximize
the score. Score optimization is what is done when, for example,
the probability of uncomplicated control is (mis)used as the score
function (see below).
Constrained Optimization
The second approach is one in
which the optimizer seeks to maximize the score
subject to
constraints on one or more measures
such as those listed above.
These constraints form a threshold above or below which a given
measure must lie.
4
As an example of
constrained optimization
, one
might seek to maximize the mean dose to the target volume subject
to the requirement that the maximum dose to each organ-at-risk of
interest is less than the predefined maximum allowed dose for that
organ.
3
It is very hard to estimate numerically, for example, the relative importance of
the mean dose to the target volume and the standard deviation of dose (a
measure of dose inhomogeneity) within the target volume. Or, how can one
know the relative importance of the D
95%
of the target volume and the V
20Gy
of
an OAR?
4
This was the approach reported by Niemierko (1992) in which the user was
allowed to pick both the measure to be used as the score function and those to
be used as constraints, from a long menu of possibilities.
