Automating Treatment Planning for Brachytherapy with Machine Learning

Important results from the project

The purpose of the visit was to develop methods that combine the use of mathematical optimization and AI and strengthen the applicant´s knowledge of AI. The focus was on two main projects, first to develop better models for the placement of needles for brachytherapy treatment of prostate cancer, and second to investigate a new type of applicator for brachytherapy treatment of rectal cancer. The overall goal of the research is to improve brachytherapy treatments of cancer tumors, with better tumor control and less risk of serious side effects from the treatment.

Expected long term effects

For project 1, AI-based methods have been used to formulate objectives that take into account the spatial distribution of the radiation dose, something that is not possible with current models. Effective such models have been developed, which contributes to automating this part of dose planning. For project 2, rotating shields are used to reduce the dose in certain directions and protect healthy tissue. The project is a study for clinical trials and aims to understand and compensate for uncertainties. Preliminary results have been presented at the Curietherapies conference.

Approach and implementation

The applicant has had a close collaboration with hospital physicists, engineers, and computer scientists with a focus on AI. Furthermore, the applicant has been part of an AI group with regular meetings and seminars. For project 1, AI-based data-driven methods have been used to be able to formulate optimization models. For project 2, models and solutions have provided important information about which uncertainties are acceptable from a clinical perspective, but have also shown that the effect of uncertainties and possible errors can be reduced by considering them in the planning.