Data science and artificial intelligence in radiation oncology

Thanks to the ever increasing role of computational power this field has advanced fast. Although no other cancer modality depends as much on modern IT technology as radiotherapy, AI applications entered radiation oncology only recently and much later compared to other medical disciplines (e.g. ophthalmology, dermatology, radiology). One pillar is to develop and later on utilize AI techniques in areas where a high level of automatization allows to implement better, faster and more individualized treatments. A second pillar is the use of deep learning techniques to provide better insight of treatment outcome. The goal is to develop better models to predict survival and toxicities for our patients. By incorporating real-world data, these models are aimed to improve our understanding of dose response relationships and help in making better treatment decisions.

Our application-oriented research and development focuses on AI techniques that are inherently aligned with the key-objectives in radiation oncology, i.e. optimizing tumor coverage, while sparing normal tissues and cover the following topics: deep learning based auto-segmentation, outcome modelling and subsequent application in knowledge based treatment planning including computerized treatment plan optimization, and the generation of synthetic CT images for MR-guided radiotherapy. A key challenge is the curation of high quality, annotated, clinical datasets which form the cornerstone for data science big data. Our team has been dedicated to establish an accessible research database within the infrastructure of the Medical University of Vienna.