Current methods of inverse planning incorporate unsophisticated decision-making components. Some of the undesirable characteristics of these methods include the inability to deal with uncertainty, the reliance on crude forms (such as weighting factors) of user preferences, and the need to proceed through numerous, cumbersome trial-and-error inverse planning attempts in order to present the planner with plans that characterize the trade-offs inherent in the decision making process. We propose to develop an improved decision making process that couples multiobjective evolutionary algorithms with influence diagrams in a multiobjective decision making environment. The hypothesis is that such an approach will result in plans that more closely reflect the planner's clinical goals, that incorporate in an explicit manner the data from clinical trials, that apply the principles of decision making under uncertainty in a way that results in more clinically acceptable plans, and that take into account the preferences of the patient and physician. Our approach is to further develop our inverse planning capabilities to efficiently search the space of possible plans for Pareto efficient plans under the multiple objectives of a case. This planning system will be coupled with an influence diagram. The influence diagram is based on a Bayesian network that incorporates expert physicians' reasoning and judgements regarding the important parameters of both the plan's 3D dose distribution and patient-related conditions. The influence diagram combines this diagram with a utility node that includes physician or patient preferences regarding the possible outcomes. We will develop and evaluate influence diagrams for prostate cancer and for head & neck cancer.
