Objective Selection for Cancer Treatment: An Inverse Optimization Approach

Abstract

The quality of radiation therapy treatment plans and the efficiency of the planning process are heavily affected by the choice of planning objectives. Although simple objectives enable efficient treatment planning, the resulting treatment quality might not be clinically acceptable; complex objectives can generate high-quality treatment, yet the planning process becomes computationally prohibitive. In “Objective Selection for Cancer Treatment: An Inverse Optimization Approach,” by integrating inverse optimization and feature selection techniques, Ajayi, Lee, and Schaefer propose a novel objective selection method that uses historical radiation therapy treatment data to infer a set of planning objectives that are tractable and parsimonious yet clinically effective. Although the objective selection problem is a large-scale bilevel mixed-integer program, the authors propose various solution approaches inspired by feature selection greedy algorithms and patient-specific anatomical characteristics.