Inverse plan optimization accounting for random geometric uncertainties with a multiple instance geometry approximation (MIGA).

Abstract

Radiotherapy treatment plans that are optimized to be highly conformal based on a static patient geometry can be degraded by setup errors and/or intratreatment motion, particularly for IMRT plans. To achieve improved plans in the face of geometrical uncertainties, direct simulation of multiple instances of the patient anatomy (to account for setup and/or motion uncertainties) is used within the inverse planning process. This multiple instance geometry approximation (MIGA) method uses two or more instances of the patient anatomy and optimizes a single beam arrangement for all instances concurrently. Each anatomical instance can represent expected extremes or a weighted distribution of geometries. The current implementation supports mapping between instances that include distortions, but this report is limited to the use of rigid body translations/ rotations. For inverse planning, the method uses beamlet dose calculations for each instance, with the resulting doses combined using a weighted sum of the results for the multiple instances. Beamlet intensities are then optimized using the inverse planning system based on the cost for the composite dose distribution. MIGA can simulate various types of geometrical uncertainties, including random setup error and intratreatment motion. A limited number of instances are necessary to simulate Gaussian-distributed errors. IMRT plans optimized using MIGA show significantly less degradation in the face of geometrical errors, and are robust to the expected (simulated) motions. Results for a complex head/neck plan involving multiple target volumes and numerous normal structures are significantly improved when the MIGA method of inverse planning is used. Inverse planning using MIGA can lead to significant improvements over the use of simple PTV volume expansions for inclusion of geometrical uncertainties into inverse planning, since it can account for the correlated motions of the entire anatomical representation. The optimized plan results reflect the differing patient geometry situations which can be important near the surface or heterogeneities. For certain clinical situations, the MIGA optimization approach can correct for a significant part of the degradation of the plan caused by the setup uncertainties.