Fraunhofer Institute for Industrial Mathematics ITWMAs part of the project »KI-ROBUST«, we are collaborating with partners from the fields of research and medicine to develop mathematical methods and AI techniques that make the planning of radiation therapy faster, safer, and more reliable. The goal is to support physicians in treating cancer patients, better account for uncertainties, and further improve the quality of treatment.
Projekt »KI-ROBUST«: Künstliche Intelligenz und Robuste Optimierung für den Umgang mit Unsicherheiten in der Strahlentherapie: Wie können wir Strahlentherapie schneller, präziser und verlässlicher machen, gerade bei steigenden Zahlen von Patientinnen und Patienten und immer komplexeren Daten?
© iStockphoto
Safe and Efficient Radiation Therapy Through AI and Robust Optimization
Project »KI-ROBUST«: Artificial Intelligence and Robust Optimization for Managing Uncertainties in Radiation Therapy
The number of cancer cases is rising, while at the same time there is a shortage of skilled personnel in hospitals. As a result, automation is becoming increasingly important in radiation therapy planning. Modern AI methods already automatically detect and highlight tumors and organs in medical images. However, uncertainties remain – such as image noise, patient movement, or unclear tissue boundaries. These uncertainties make it difficult to use automated methods in clinical practice.
Making AI Results Easy to Understand and Follow
Together with our project partners, we are developing methods that make the decisions of AI systems transparent. The AI should not only suggest a contour for tumors or sensitive organs, but also be able to explain why it arrived at that result.
In addition, the models automatically highlight areas where uncertainty is particularly high. This allows doctors to focus specifically on critical regions instead of having to manually review all contours. This saves time while also increasing the safety of treatment.
Our Expertise: Robust Optimization for Better Treatment Plans
Our team develops mathematical methods for robust optimization. This results in radiation treatment plans that function reliably even when uncertainties arise in the underlying data.
Our models specifically analyze which uncertainties have a particularly strong influence on the quality of a treatment plan. Based on this analysis, we develop robust treatment plans that strike a sensible balance between treatment quality and safety. At the same time, we provide transparency regarding how different decisions affect the robustness of the planning.
Inverse-robust multi-criteria optimization of radiation treatment plans based on uncertainty-aware, explainable AI-based vehicle segmentation models in adaptive radiation therapy.
Take Into Account Uncertainties Throughout the Entire Treatment
The treatment of cancer patients often spans several radiation therapy sessions. During this time, tumors and organs may change or shift. Such changes affect the actual radiation dose administered.
We are therefore developing methods that incorporate previously administered doses – as well as the uncertainties associated with them – into further planning. This allows radiation treatment plans to be continuously adapted to new circumstances. The goal is to ensure treatment that is as precise and safe as possible throughout the entire course of therapy.
Intelligently Combining AI and Mathematical Optimization
Calculating robust treatment plans requires complex optimization methods. That is why we are exploring new approaches that combine Artificial Intelligence with mathematical optimization methods.
This combination significantly speeds up calculations without compromising the quality of the results. At the same time, we are laying the groundwork for explainable and reliable decision support in clinical applications.
Outlook and Benefits for Patients
The project contributes to more efficient and precise cancer treatment. Transparent AI methods and robust optimization help to better manage uncertainties and make treatment plans more reliable.
In the long term, the methods developed are intended to be incorporated into clinical software solutions. As a result, more patients could benefit from high-quality, personalized radiation therapy in the future.
Our Partners in Research, Healthcare, and Industry
The collaborative project brings together expertise in mathematics, artificial intelligence, and radiation therapy.
Time-efficient, robust optimization of adaptive radiation therapy plans that takes accumulated dose into account, using AI-based uncertainty estimation.
- Ludwig Maximilian University of Munich (MFAI) contributes its expertise in the mathematical foundations of Artificial Intelligence and develops methods for reliable and explainable AI systems.
- LMU Medical Center Munich, Adaptive Radiotherapy Lab is conducting research on AI-based methods for image-guided adaptive radiation therapy and evaluating the developed methods using clinical data.
- Brainlab is supporting the project as an application partner and contributing its expertise in clinical software solutions for radiation therapy.
- Varian Medical Systems, a Siemens Healthineers company, is supporting the project with its expertise in the development and implementation of state-of-the-art radiation planning systems.
Project Funding and Duration
The project is funded by the Federal Ministry of Education, Research, and Technology (BMFTR) through the »Mathematics for Innovation« program. It is scheduled to run from February 1, 2026, to January 31, 2029.