Robust Multi-Criteria Optimization With Application to Radiation Therapy

Northwestern University At Chicago, Evanston IL

Investigators

Abstract

The research goal of this award is to significantly expand the range of applicability of multi-criteria optimization when input data are uncertain and may exhibit correlations. This includes the collection and analysis of potentially erroneous data, the research of robustness in correlated multi-expert multi-criteria problems, and the validation of the methods on available approaches and probing of results on real data. Many real-world applications put a decision maker in the position to simultaneously achieve multiple contradicting goals. The most common approach is to minimize an aggregate of the objectives, each of which are given a positive weight by its importance. Often, these weights are determined by one or more experts, leading to a sizable level of uncertainty. The results from this research will advance the incorporation of experts' preferences into decision- making and the reliability and reproducibility of decisions. The generality of the approach makes it broadly relevant to real-world problems, many with objectives that are naturally correlated. In radiation therapy, the approach enables oncologists' choices to directly inform the method. Human bias in clinical decisions will be minimized, making optimal cancer treatment accessible, relevant, and ultimately beneficial to patients regardless of medical facility or geographic region. From an educational standpoint, the direct interaction between engineering students and clinicians will offer a unique training ground for the next generation of cross-disciplinary researchers. Currently, the two main paths in multi-criteria optimization in the presence of errors are Pareto frontier approaches that do not produce one final decision and distributionally robust approaches that are limited to the availability of the distribution. This project focuses on an extension of the latter and is motivated by the application in radiation therapy optimization, as used in nearly two thirds of all cancer cases. This inverse problem typically follows a trial-and-error procedure. The research agenda provides deeper insights on a number of levels into fundamental questions in robust multi-criteria decision-making problems, as well as to establish tools and solutions directly relevant to clinical practice. This project, if successful, can lead to a drastic reduction of both decision time and ambiguity of the outcome, warranting high-quality cancer radiation treatments not limited by human and institutional uncertainties.