On Quality in Radiotherapy Treatment Plan Optimisation

Radiotherapy is one of the essential treatments used in the fight against cancer. The goal of radiotherapy is to deliver a high dose of ionising radiation to the tumour volume and at the same time minimise the effect on healthy tissue by reducing the radiation to critical organs. This contradiction is challenging and has been driving the research and development of the treatments.
Over the last two decades, there has been tremendous technical development in
radiotherapy. The rapid increase in computational power introduced treatment plan optimisation and intensity-modulated radiotherapy (IMRT). IMRT made it possible to shape the radiation dose distribution closely around the target volume avoiding critical organs to a greater extent. Rotational implementation of IMRT, e.g. Volumetric Modulated Arc Therapy (VMAT) further improved this “dose shaping” ability. With these techniques increasing the ability to produce better treatment plans, there was a need for evaluation tools to compare the treatment plan quality. A plan can be judged by how well it fulfils the prescription and dose-volume constraints, ideally based on treatment outcome. In this work, this is denoted Required Plan Quality, the minimum quality to accept a plan for clinical treatment. If a plan does not fulfil all the dose-volume constraints, there should be a clear priority of which constraints are crucial to achieve. On the other hand, if the constraints are easily fulfilled, there might be a plan of better quality only limited by the treatment systems ability to find and deliver it. This is denoted Attainable Plan Quality in this work– the quality possible to achieve with a given treatment system for a specific patient group.
In work described in this thesis, the so-called Pareto front method was used to search for the attainable plan quality to compare different treatment planning systems and optimisation strategies. More specifically, a fall-back planning system for backup planning and an optimiser to find the best possible beam angles. The Pareto method utilises a set of plans to explore the trade-off between target and nearby risk organs.The Pareto plan generation is time-consuming if done manually. The Pareto method was then used in a software that automated the plan generation allowing for a more accurate representation of the trade-off. The software was used to investigate the attainable plan quality for prostate cancer treatments. In the last two publications in this thesis, machine learning approaches were developed to predict a treatment plan
closer to the attainable plan quality compared to a manually generated plan.

In the thesis, tools have been developed to help move the treatment plan quality
from Required Plan Quality towards the Attainable Plan Quality, i.e. the best quality we can achieve with our current system.