Premium
SU‐FF‐T‐256: IMRT Inverse Planning with Voxel‐Based Penalty Scheme
Author(s) -
Callaghan T,
Levy D,
Xing L
Publication year - 2007
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.2760917
Subject(s) - voxel , mathematical optimization , inverse problem , weighting , inverse , algorithm , mathematics , penalty method , computer science , artificial intelligence , mathematical analysis , physics , geometry , acoustics
Purpose: A tacit assumption made in current inverse planning algorithms is that all points within a structure (target or sensitive organ) are equivalent, as reflected by the fact that the dose prescription, weighting factors and dose‐volume constraints are generally imposed a priori on a structure‐specific basis. This seriously limits the IMRT solution space and compromises the achievable dose distributions. The purpose of this study is to develop a novel IMRT inverse planning framework with a more general voxel‐specific penalty. Methods: Different from conventional inverse planning in which the same penalty (objective) function is used for all voxels, here we assign each voxel a penalty proportional to[ D c( i ) − D 0( i ) ]n iwhere D c and D 0 are computed and prescribed doses, respectively, and n i is a voxel‐specific exponent. In this way, we “break” the implicit constraint that a structure is a mathematically uniform entity with all voxels having the same “identity”. The seemingly insolvable task of determining a large number of voxel‐specific { n i } is dealt by heuristically relating them to the doses they receive. The optimization starts from a uniform value of n i = 2 and, during the iterative optimization process, the n i 's are treated as parts of the system variables and are updated together with the beamlet weights. If a voxel is overdosed, for example, it means that the voxel is not penalized sufficiently and the algorithm will automatically increase the value of n i , and vice versa . Results: A voxel‐dependent penalty scheme is established for IMRT inverse planning. The formalism properly models the intra‐organ tradeoff and allows the optimizer to access solutions that would otherwise be inaccessible. Our phantom and clinical head‐and‐neck case studies show that much superior treatment plans can be obtained as compared with the conventional approaches. Conclusion: Voxel‐based penalty scheme leads to substantially improved IMRT dose distributions and allows us to maximally utilize the IMRT technology.