Purpose: To describe the theoretical basis, the algorithm, and implementati
on of a tool that optimizes segment shapes and weights for step-and-shoot i
ntensity-modulated radiation therapy delivered by multileaf collimators.
Methods and Materials: The tool, called SOWAT (Segment Outline and Weight A
dapting Tool) is applied to a set of segments, segment weights, and corresp
onding dose distribution, computed by an external dose computation engine.
SOWAT evaluates the effects of changing the position of each collimating le
af of each segment on an objective function, as follows. Changing a leaf po
sition causes a change in the segment-specific dose matrix, which is calcul
ated by a fast dose computation algorithm. A weighted sum of all segment-sp
ecific dose matrices provides the dose distribution and allows computation
of the value of the objective function. Only leaf position changes that com
ply with the multileaf collimator constraints are evaluated. Leaf position
changes that tend to decrease the value of the objective function are retai
ned. After several possible positions have been evaluated for all collimati
ng leaves of all segments, an external dose engine recomputes the dose dist
ribution, based on the adapted leaf positions and weights. The plan is eval
uated. If the plan is accepted, a segment sequencer is used to make the pre
scription files for the treatment machine. Otherwise, the user can restart
SOWAT using the new set of segments, segment weights, and corresponding dos
e distribution. The implementation was illustrated using two example cases.
The first example is a T1N0M0 supraglottic cancer case that was distribute
d as a multicenter planning exercise by investigators from Rotterdam, The N
etherlands. The exercise involved a two-phase plan. Phase I involved the de
livery of 46 Gy to a concave-shaped planning target volume (PTV) consisting
of the primary tumor volume and the elective lymph nodal regions II-IV on
both sides of the neck. Phase 2 involved a boost of 24 Gy to the primary tu
mor region only. SOWAT was applied to the Phase 1 plan. Parotid sparing was
a planning goal. The second implementation example is an ethmoid sinus can
cer case, planned with the intent of bilateral visus sparing. The median PT
V prescription dose was 70 Gy with a maximum dose constraint to the optic p
athway structures of 60 Gy.
Results: The initial set of segments, segment weights, and corresponding do
se distribution were obtained, respectively, by an anatomy-based segmentati
on tool, a segment weight optimization tool, and a differential scatter-air
ratio dose computation algorithm as external dose engine. For the supraglo
ttic case, this resulted in a plan that proved to be comparable to the plan
s obtained at the other institutes by forward or inverse planning technique
s. After using SOWAT, the minimum PTV dose and PTV dose homogeneity increas
ed; the maximum dose to the spinal cord decreased from 38 Gy to 32 Gy. The
left parotid mean dose decreased from 22 Gy to 19 Gy and the right parotid
mean dose from 20 to 18 Gy. For the ethmoid sinus case, the target homogene
ity increased by leaf position optimization, together with a better sparing
of the optical tracts.
Conclusions: By using SOWAT, the plans improved with respect to all plan ev
aluation end points. Compliance with the multileaf collimator constraints i
s guaranteed. The treatment delivery time remains almost unchanged, because
no additional segments are created. (C) 2001 Elsevier Science Inc.