Data driven decision support for more efficient radiotherapy

Important results from the project

The project largely achieved its goals. An AI model for tumor contouring was created using transfer learning. A parallel optimization algorithm (ADMM) enables generation of hundreds of Pareto-optimal plans in under two minutes. A method for Pareto navigation with a GUI was developed, including patenting. An integrated prototype was demonstrated to a clinical expert group with positive feedback after revision. Two articles were published in Physics in Medicine & Biology and Medical Physics.

Expected long term effects

The project´s results can contribute to faster, more consistent radiotherapy planning. ADMM-based Pareto generation enables multicriteria optimization within clinical timeframes, potentially reducing variability and operator dependency. Automated contouring and Pareto navigation can free up clinical time and make high-quality planning accessible at more clinics. The patent application and scientific publications lay the groundwork for future product development and commercialization at Elekta.

Approach and implementation

The project was a 3-year collaboration between Elekta (coordinator) and Uppsala University, with 3 subprojects: AI-based contouring, Pareto generation and Pareto navigation. UU´s postdoctoral fellow worked regularly at Elekta´s office, and there were physical meetings for the project group. There was an initial delay due to contractual issues and a residence permit issue, which was picked up through an updated schedule. The project was extended by 4 months due to parental leave.