Mikolaj Stryja

Research Objectives

Radiotherapy (RT) is one of the most common ways to treat cancer in clinical practice. To further improve its efficacy and increase patient quality of life after treatment, the dose in Organs at Risks (OARs) must be reduced as much as possible, without sacrificing tumor coverage. While adapting treatments to anatomical changes (such as tumor shrinkage or weight loss) is becoming increasingly possible, key algorithmic challenges in several steps of the treatment planning workflow still prevent the large-scale adoption of fully automated, robust daily – and ultimately real-time - adaptive treatments that would maximally minimize OAR doses.

Artificial Intelligence (AI) based solutions could be crucial to overcoming these challenges and support clinicians in their daily practice. Thanks to the increased availability of computational resources and structured datasets, recent deep learning models showed state-of-the-art performance in many RT tasks, such as contouring or predicting achievable dose distributions. Several unsolved challenges remain however, including the optimal utilization of multimodal imaging, biological data and patient monitoring information; the fast prediction of clinically optimal robust doses together with their corresponding plan and machine parameters; biology guided adaptation strategies driven by outcome models; or the inclusion of uncertainties in AI models. 

As a member of the Biological Intervention Optimisation AI Lab (BIOLab), the aim of my research is to develop novel machine and deep learning methods in order to overcome some of these challenges and further personalize radiotherapy treatments, particularly by using Generative Adversarial Networks (GANs) and Reinforcement Learning (RL) approaches. By focusing on both algorithmic innovation and practical implementation aspects, I believe my research will contribute to more accurate and effective radiotherapy treatments, ultimately having a positive impact on patients’ lives.