TRiP98 optimize command
optimize / iterations() graceiterations(1) H2Obased CTbased events() debug
bio singly eps(1E-3) geps(2E-3) myfac(0.5)
phys matchonly doseweightfactor()
field(*) nopreopt
grad4zero
optalgorithm(fr)
bioalgorithm(ld)
dosealgorithm(allpoints)
Purpose
Handles optimization of dose with respect to particle numbers.
Parameters
- iter()
-
Maximum number of iterations. Remembered.
Startup default is 5.
- graceiter()
-
Number of initial "grace" iterations
for which tests on relative change of chi-square
are suspended.
Startup default is 1.
- bio
-
Optimize using biological dose. Remembered.
- phys
-
Optimize using physical dose. Remembered.
- H2Obased
-
- CTbased
-
Optimization principle.
H2Obased
selects separate optimization
of single fields, with fast, but less accurate algorithms,
based on a H2O-equivalent grid. This is the default.
CTbased
selects optimization based on the
CT grid. Mandatory for simultaneous optimization of
multiple fields.
- singly
-
Separate optimization of multiple fields,
i.e. combination of two separate optimization calls into one.
Note, however, that the target dose is still
partitioned between the multiple fields.
- nopreopt
-
Skips H2Obased pre-optimization
so CTbased is invoked with zero particle number
as startup value. Use it with care.
- matchonly
-
Optimize using only a simple peak-to-dose matching algorithm.
Otherwise a conjugate gradient method is used to equilibrate
peaks and valleys. This applies to
H2Obased optimizations only.
- grad4zero
-
For optimizations based on gradients of the objective function:
include also contributions from raster points with zero particle count.
By default, such points are ignored (except for nopreopt).
As a side-effect, this
permanently excludes raster positions from the optimization
one their particle numbers are below the
minparticle() threshold.
grad4zero
counteracts this, if desired.
- dosealgorithm(dalg)
-
Selects
algorithm
for dose calculation.
Same choices as for
dose cube calculation:
-
classic
or cl
(default)
-
allpoints
or ap
-
multiplescatter
or ms
- bioalgorithm(balg)
-
Selects
algorithm
for biological effect calculation.
-
classic
or cl
-
lowdose
or ld
(default)
Note, however, that the default choice is valid only for doses
up to some 10GyE, but is considerably faster.
- optalgorithm(oalg)
-
Selects
algorithm
for dose optimization.
The selection is affects
CTbased
optimizations, as of version 1001d H2Obased too. Choices are
-
classic
or cl
(classic)
-
cjggrad
or cg
(conjugate gradients)
-
gradient
or gr
(plain gradients)
-
bortfeld
or bf
(Bortfeld's algorithm)
-
fletcher-reeves
or fr
(Fletcher-Reeves algorithm, default)
-
bfgs
or bfgs(angle,armdamp,armdelta)
:
BFGS algorithm, with its crucial parameters
angle
(angle test criterion, default=0.2),
armdamp
(Armijo damping factor, default=0.5),
armdelta
(Armijo delta factor, default=0.01).
As it turns out, "Fletcher-Reeves" is usually the fastest converging one.
BFGS comes second, however, it needs considerably more RAM.
- events(nevt)
-
Not yet implemented.
- eps(eps)
-
Minimum relative change of chi-square.
If the actual change is below, the iteration is aborted.
Startup default is 1E-3.
- geps(geps)
-
Controls relative contribution of raster points to dose voxels.
Contributions less than
geps
are omitted,
thereby reducing memory and time consumption.
Should not be raised above some 1E-3, otherwise optimized
and real dose distributions will differ.
Startup default is 2E-3.
- myfac(myfac)
-
Controls search stepsize for gradient based
optimization algorithms for the RBE-weighted (biological) dose.
The default value is 0.5, designed for total biological target doses about 3GyE.
For lower doses (0.5..1GyE) or high RBE, lower values of
myfac
might be appropriate.
There is some builtin stepsize "autopilot", but if the results aren't
satisfactory, one might manually choose lower values, e.g.
down to 0.1 to 0.25.
As of version 1001D, this parameter applies for
H2Obased optimizations as well.
- doseweightfactor(dwf)
-
The prescribed dose is multiplied with this factor to introduce sort of
an "experimental uncertainty", which enters as a weight in the chi-square
evaluation.
Startup default is 2.5E-2, corresponding to half of the
deviation between measured and calculated absorbed dose allowed by authorities.
- field(idlist)
-
List of field id's to be optimized.
- debug
-
Debug switch. Lots of output!
Remarks
-
Not all optimization algorithms can cope with all scenarios
equally well. Sometimes choosing another algorithm might help.
Examples
-
opt / field(*) bio
Optimize all defined fields using the biologically effective dose.
Fields are optimized separately with "classic" algorithms
-
opt / field(*) ctbased bio dosealg(ap) optalg(fr) bioalg(ld) geps(2E-3) eps(1e-2) iter(500)
Optimize all defined fields simultaneously (IMRT-style),
with "fast" biological calculations and the Fletcher-Reeves algorithm.
Last updated:
$Id: trip98cmdopt.html,v 1.11 2012/06/01 18:48:43 kraemer Exp $
M.Kraemer@gsi.de