The goal of the Bavarian Cancer Research Center (BZKF) is to provide cancer patients in Bavaria with access to cutting-edge oncology care close to home, offering the best possible treatment based on the latest scientific evidence. The BZKF Lighthouse for AI & Bioinformatics brings together staff from all six university hospital sites in Bavaria, creating a comprehensive oncology research infrastructure. Its mission is to analyze oncological real-world data (RWD) across institutions and to generate new insights into cancer treatment and prognosis through the application of AI methods.
With the BAIOSPHERE x SSC program, we would like to seize the opportunity to organize an intensive working and exchange retreat, bringing together clinical experts, AI researchers, data scientists, and early-career investigators from all six BZKF sites. Over five intensive days, we will pool expertise, data, and computational innovation to push the boundaries of what is currently possible in AI-driven oncology research.
A key objective of the retreat is to advance existing datasets and perform in-depth analysis with preliminary promising results towards high-impact publications. We plan on doing so by combining breakout sessions in smaller groups based on specific research questions with shared, collaborative meetings. Two major research questions are to be addressed:
Across the sites, decades of clinical data on conditions, procedures, and treatments for cancer patients have previously been harmonized and integrated. This real-world data contains records of more than 300 000 patients across the six university hospitals. Based on this data, we want to develop AI-based trajectory models that capture the sequence and timing of diagnostics, treatments, and outcomes across multiple institutions. The goal is to identify common treatment pathways, deviations, and their associations with patient outcomes.
As an initial application, we will focus on younger prostate cancer patients (<65 years old) with aggressive Gleason scores. For this subgroup, single-center data is insufficient due to low case numbers. By pooling data across all sites, we can develop prognostic models for treatment response and survival that are both statistically robust and clinically meaningful.
Further, the model architecture and analytical pipeline will be designed to be entity-agnostic, allowing adaptation to other malignancies. This ensures that the methods developed during the retreat can be scaled across the entire oncology spectrum, enabling multi-disease, multi-modal predictive modeling on real-world data at scale.
The second available dataset includes 3 000 colon cancer patients with detailed histopathological annotations. It offers a unique opportunity for developing cutting-edge AI models for cancer classification, grading, and outcome prediction. Beyond the single dataset we also want to develop a project proposal which combines the detailed histopathology data with the granular clinical data from the first research question – laying the foundation for precision oncology at scale.
Please note: This is just an information regarding events taking place at SSC; public attendance is therefore not possible due to the character of the retreat.
