Designing Digital Twins for Personalized Risk Assessment in Breast Cancer Radiotherapy: Insights From the Tetris Project
Please login to view abstract download link
We spotlight the activity of the Tetris project that aims to improve radiation protection and patient safety in breast cancer radiotherapy by developing digital twin (DT) technology. The project focuses on creating predictive models that integrate various data types to enable personalized assessments of long-term risks. In precision oncology, DTs should be seen as comprehensive models that synthesize multiple data sources to support predictions and clinical decision-making. The primary goal is to design DTs that incorporate all relevant information while providing personalized insights. The core concept behind the DT architecture in Tetris is the interaction between a “world model” and an “agent.” The “world model” is a low-dimensional, possibly hierarchical and dynamic representation of the patient’s and disease’s states, possibly updated by integrating genomic, transcriptomic, imaging, and clinical data. The “agent” interacts with this model to make decisions, such as risk assessments, and to suggest actions, such as optimizing treatment or follow-up schedules. This architecture allows the integration of both retrospective population data and individual patient data, leveraging the hierarchical structure of the world model to improve predictive accuracy. Following these general principles, we will discuss how to address three key activities. The first one is the development of prototype digital twins from retrospective data. Initial digital twins are being developed using data from the REQUITE cohort (~2000 patients). These twins will incorporate patient-specific anatomical and clinical data, along with treatment details, to generate personalized risk scores and model the temporal dynamics of severe side effects post-RT. The second one is about the refinement of digital twins using prospective data. High-fidelity digital twins will be refined using more comprehensive data from the Tetris cohort (~250 patients), including baseline biomarkers, genetic data, environmental factors, and lifestyle information. These refined models will enable a deeper, more accurate representation of the patient’s health and risk profile over time. Finally, we will discuss perspectives on applying the refined digital twin approach to provide personalized follow-up schedules.
