Measuring radiation treatment plan similarity in the cloud

Abstract

Radiation therapy is a form of cancer treatment in which carefully designed plans are used to direct treatment over multiple occasions (fractions). Creating radiation plans is quite laborious, so it is not feasible to manually create a plan for each fraction to maintain treatment quality. We propose to use a database of plans to find the most similar anatomy, based on which a suitable daily plan might be automatically created, thus reducing staff time. However, the computation for finding the most similar plan is long and computationally intensive. We present a method for finding the most similar plan using cloud resources to reduce computation time. The preliminary, unoptimized implementation finds the most similar plan in less than 7 minutes when choosing from five plans, and it is anticipated that increasing the number of plans will result in only a small relative increase in computation time.