Abstract:
Radiation therapy is a main cancer treatment modality that exploits ionizing radiation to deliver a therapeutic dose to tumorous tissues in order to sterilize the proliferation of clonogenic cells. Advanced high-precision conformal radiation therapy techniques like intensity-modulated radiotherapy (IMRT) offer technical solutions to deposit a sufficiently high dose to the tumor and simultaneously spare surrounding healthy tissues as much as possible. The inherent technical complexity of IMRT requires an inverse treatment planning approach, which relies on mathematical optimization techniques to search for an optimal treatment plan that satisfies the clinical goals. The problem is that these goals are mutually dependent and contradictory, and therefore a single best solution does not exist in general. Instead, the large number of degrees of freedom of IMRT to shape the dose distribution has opened the possibility to generate multiple solutions and to allow physician/patient specific trade-offs between achieving local tumor control and causing severe side effects in surrounding normal tissues. Therefore, there is a need for methods to find best-possible plans and assist in (interactive) balancing of patient- and physician-specific preferences.
This lecture will address methods from the field of Operations Research to deal with the multi-objective optimization problem of IMRT planning. Finding the best-possible compromise solutions requires user interaction and guidance from the search algorithm. Different methods to articulate preference information in inverse treatment planning will be reviewed. A priori preference methods use weighting factors to capture the relative importance of the objectives, but generally require several trial and error iterations to find an acceptable plan, because the relationship between the goals is unknown beforehand. Furthermore, the plain weighting factors do not reflect clinical importance, and the sensitivity to changes in the weighting factors is unknown. Therefore, in clinical practice the potential of IMRT cannot be fully exploited. To overcome the time-consuming human iteration loop, a posteriori preference methods for off-line generation of best-compromise solutions have been applied and explored. First, the set of Pareto solutions is identified for which improving one objective cannot be improved by deteriorating at least one other objective, and then the best solution is selected from a graphical representation of this set (Pareto front) according to individual preferences. In this way the treatment planner/physician can experience the sensitivity to changes in the goals and decide for the clinically optimal compromise.
The huge number of best-possible compromises requires: 1) efficient algorithms to approximate the Pareto front and 2) support tools for systematic interactive decision-making. Examples of these will be discussed and practical aspects relating to the use of these algorithms and tools will be elucidated. The feasibility of using Pareto fronts to quantitatively compare treatment plans between different treatment planning systems and treatment techniques has recently been described and will be addressed in this lecture.