Decision theoretic steering and genetic algorithm optimization: Application to stereotactic radiosurgery treatment planning

Publisher Website
Google Scholar
Abstract
Treatment planning for stereotactic radiosurgery and fractionated radiotherapy is currently a labor intensive, operator‐dependent process. Many degrees of freedom exist to make rigorous optimization intractable except by computationally intelligent techniques. The quality of a given plan is determined by an aggregate of clinical objectives, most of which are subject to competing tradeoffs. In this work, we present an autonomous scheme that couples decision theoretic guidance with a genetic algorithm for optimization. Ordinal ranking among a population of viable treatment plans is based on a generalized distance metric, which promotes a decreasing hyperfrontier of the efficient solution set. The solution set is driven toward efficiency by the genetic algorithm, which uses the tournament selection mechanism based on the ordinal ranking. Goals and satisficing conditions can be defined to signal the ultimate and the minimum achievement levels in a given objective. A conventionally challenging case in radiosurgery was used to demonstrate the practical utility and the problem‐solving power of the decision theoretic genetic algorithm. Treatment plans with one isocenter and four isocenters were derived under the autonomous scheme and compared to the actual treatment plan manually optimized by the expert planner. Quality assessment based on dose–volume histograms and normal tissue complication probabilities suggested that computational optimization could be driven to offer varying degrees of dosimetric improvement over a human‐guided optimization effort. Furthermore, it was possible to achieve a high degree of isodose conformity to the target volume in computational optimization by increasing the degree of freedom in the treatment parameters. The time taken to derive an efficient planning solution was comparable and usually shorter than in the manual planning process, and can be scaled down almost linearly with the number of processors. Overall, the autonomous genetic algorithm scheme was found to be powerful and versatile as a computationally intelligent counterpart to human‐guided strategies in treatment optimization for stereotactic radiosurgery and radiotherapy.