"Best Objective Selection in Multiobjective Optimization"
Radiation therapy is a common method of cancer treatment. A challenge in radiation therapy treatment planning is selecting which clinical objectives to use in the optimization and determining their relative importance. We propose the inverse optimization method with a cardinality constraint to infer the most important objectives from historical treatment plans. We use a greedy algorithm to select objectives and provide theory, a generalization of a result by Nemhauser et al. (1978), to support our results. We compare the proposed method to the cardinality-constrained inverse problem and show that our method efficiently finds a small subset of objectives that generates clinical quality treatment plans.