Description

Cancer is a major threat to human health and development (World Health Organization). New cases for high income countries in 2020 are expected to be 21% higher than in 2009 (Economist Intelligence Unit). The European Commission (EC) considers that cancer “represents an enormous burden on society in an ageing Europe, affecting a growing number of individuals and their relatives”, and recognizes that “Europe is characterized by worrying inequalities in cancer control and care existing within, as well as between, EU Member states”. According to EC, about 50% of all cancer patients are treated with radiotherapy during the illness.

The current clinical planning practice creates huge variations in the quality of the delivered treatments, since treatment quality is dependent on the planner’s skills and time availability. It is not possible to know how far from the optimal plan the current plan is and it can be unbearable to spend several days planning the treatment for a single patient. Some Treatment Planning Systems (TPS) make available some level of automated planning, supporting constrained optimization, sensitivity analysis, exploration of trade-offs, optimization based on templates. Nevertheless, there is no TPS on the market that releases the human planner from all the tuning and trial-and-error procedures: there is no fully automated treatment planning software solution.

Treatment dose is usually fractionated in daily sessions, during several weeks. If there are considerable changes on the patient’s anatomy or in the size and position of the areas to treat, then re-planning has to be considered. The final objective is to have daily re-planning (Adaptive Radiotherapy - ART), which will require also new developments in autosegmentation and deformable image registration tools, together with automated treatment planning.

Nowadays there are many different treatment modalities available, and new radiotherapy treatment approaches will probably appear in the future. The lack of automated treatment planning solutions prevents the possibility of taking full advantage of the existing treatment modalities: it is not possible to know how far it would be possible to go regarding treatment quality with a given technology. Furthermore, treating a patient with a more recent modality can be more expensive than using a conventional one. Deciding on the best treatment has to consider the tradeoff between costs and benefits to the patient of using a more expensive treatment. The treatment plan is patient and machine dependent. It is thus necessary to generate different plans for different modalities to support the decision making process, which is currently not possible due to the overload that would bring to an already heavy workflow.

Automated planning will contribute to: personalized medicine in radiotherapy; increase in treatment’s quality with expected increase in disease control and reduced morbidity; treatment practice uniformity between different clinicians and institutions; better selection of patients for new and more expensive treatment modalities; developments in ART; significant decrease in costs. These impacts will have important social and economic consequences, due to their influence at patient and family levels and work productivity.

The project is aligned with the European Society for Radiotherapy & Oncology Vision 2020: “Every cancer patient in Europe will have access to state of the art radiation therapy, as part of a multidisciplinary approach where treatment is individualized for the specific patient’s cancer, taking account of the patient’s personal circumstances.” It is also aligned with key objectives of the EU Framework for Research and Innovation Horizon 2020 (Excellence Science, Competitive Industries, Better Society) and with the Portuguese National Research and Innovation Strategy for Smart Specialization. It will provide training, research and development career opportunities for young researchers, involving the scientific community in economic activities and contributing to scientific job creation. “Information Technology” and “Health and Well Being” are thematic differentiated domains for Portuguese Centre Region. One of the priority areas connecting the thematic domains is “Technologies for Quality of Life”, focusing on personalized medicine, consolidation of excellence in clinical practice.

In this project IMRT as well as VMAT (where radiation arcs instead of discrete set of radiation directions are used) will be considered. Regarding IMRT BAO, the focus will be on non-coplanar treatments, since it is where the most important added-value of BAO is found. Non-coplanar BAO is seldom done in clinical practice due to the increased level of complexity related to the increased levels of freedom introduced in the problem.

Non-coplanar IMRT BAO problem will be interpreted as a continuous non-convex optimization problem (whilst other approaches known from the literature treat this as a combinatorial problem), to be tackled by derivative-free algorithms, population based and local search based metaheuristics. Parallel strategies for implementing these optimization methods will have to be devised. Although methods developed for IMRT planning cannot be directly applied to VMAT, VMAT planning optimization can be interpreted as IMRT planning optimization with a huge number of radiation incidences (also known as control points – CP – determined by the discretization of the arc), and with one single MLC aperture in each CP. The main advantage of VMAT seems to be the reduced time of treatment, with equal or better dosimetric results. Nevertheless, intra-motion problems can be better tackled by fixed beam methods, and dosimetric differences may not present significant clinical benefits. BAO is still used for VMAT especially considering non-coplanar treatment planning.

Generating a treatment plan is a highly complex task, involving compromises between different conflicting objectives and human judgement. The mutual dependences that exist between objectives are not known a priori and are patient dependent. We will make explicit use of conflicting objectives when guiding the algorithmic search processes. The optimized treatment plans will have to be evaluated and compared, through the development of new evaluation tools that will support the clinical decision. This will be achieved by combining the development of efficient multiobjective non-linear optimization algorithms with new methodologies for decision making in radiotherapy treatment selection. Indeed, since several contradicting objectives have to be simultaneously considered (minimization of the irradiation of healthy organs and maximization of the irradiation of the volumes to treat, for example), the optimization process will typically produce a number of alternative plans from which the final treatment plan must be selected. Moreover, attributes beyond dosimetry ones can be considered (like the patient’s preferences). Automated recommendation systems capable of suggesting one treatment plan out of the set of possible treatment plans will be developed based on preference learning and multi-attribute decision analysis approaches, while parallel computing approaches will be used in order to drastically reduce computation times and make the automated processes compatible with clinical time frames.