Biomedical Engineering Reference
In-Depth Information
Calculation of uncertainty
There are numerous sources of uncertainty in the estimation of the
dose delivered to the patient, many of which have been touched upon
in preceding chapters. Chapter 2 will surely have convinced you of
the necessity of assessing, displaying, and recording uncertainties and
the confidence level at which they have been estimated. If you are
not yet convinced, let me give you a scenario that highlights the
problem that arises if one does not analyze uncertainties. Imagine a
clinician has set a dose constraint that the center of the spinal cord not
receive more than 48 Gy and that the planner has developed a plan in
which the dose to the center of the cord is precisely 48 Gy. Upon
seeing this, an unwary clinician would probably be satisfied and
would sign off on the plan provided, of course, that the other
constraints were also satisfied. Now, suppose the planner was a little
bit savvy about uncertainties and was to warn the clinician that, yes,
the best estimate of the cord dose was indeed 48 Gy, but that there
was a 50% chance that the cord dose was higher than that (as is,
indeed, the case). The clinician's attitude towards the plan would
almost certainly change. He or she would want to know how much
over 48 Gy (at some confidence level)
5
the dose could be before
agreeing to the plan. That is, he or she would want to know the upper
bound on the dose estimate.
Although quite some attention has been given to specific sources of
uncertainty, such as patient and organ motion as discussed in Chapter
7, there has been little done to quantify the overall uncertainty in the
dose delivered throughout the patient. I have proposed a simple
approach to this problem (Goitein, 1985), which involves computing
three dose distributions, namely the nominal, upper-bound, and
lower-bound dose distributions. The “nominal” dose distribution is
based on the best estimate of all factors involved in computing the
dose. The “upper-bound” dose distribution uses extreme values (at a
specified confidence level
−
I tend to use 85% as mentioned in
5
I once had the chance to meet the US president's science advisor who, in
conversation unrelated to the reason for my visit, was bemoaning his
difficulties in presenting members of congress with an analysis of the
safety of a space mission being contemplated at the time. “They don't
want to hear that there is only one chance in ten million of a problem.
They want to know: is it safe, or not?” We in the world of radiation
oncology cannot hold out for such certainty; we must learn to be
comfortable with probabilities.
