Improving risk stratification and treatment planning in brain metastases patients: Development and validation of AI models in the AURORA multi-center patient cohort

• Applicants: Privatdozent Dr. Jan C. Peeken; Professor Daniel Rückert, Ph.D.; Privatdozent Dr. Benedikt Wiestler
• Subject Area: Nuclear Medicine, Radiotherapy, Radiobiology
• Term: since 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 504320104

In this proposal we aim to develop an artificial intelligence (AI)-based system for individualized risk assessment and target volume definition for patients with brain metastases (BM) that received primary surgical resection. To this end, we have defined five work packages addressing key research questions necessary to achieve this goal: (i) Initially, we will design and implement an AI-based, combined preprocessing and tumor segmentation pipeline using the unique multicenter retrospective “AURORA” cohort of BM patients. (ii) We will develop personalised prediction models for local and distant brain control using state-of-the-art AI techniques and radiomics of pre-surgery MRI data. Predicting these endpoints is an important step towards personalized treatment of BM patients. The incremental value in prognostic assessment conferred by advanced geometric deep learning approaches (including graph convolutional networks) integrating clinical and biological information with imaging data will be assessed. (iii) We will develop a novel framework for longitudinal analysis using pre- and post-surgical MRI scans and we will assess the added value of these longitudinal information for prognostic assessment of BM patients. (iv) Post-operative MRI and CT data will be annotated to develop an AI-based modality-agnostic pipeline for automated segmentation of resection cavities and relevant anatomical structures, ultimately, yielding an automated radiotherapy target volume definition. This pipeline could standardize the segmentation procedure and reduce the task-related working time for radiation oncologists. (v) Finally, we will integrate the developed techniques and models into a clinical decision support system that will enable individualized risk-stratification and automatized radiotherapy target volume definition. The framework will be externally validated using the AURORA trial multicenter patient data. All developled AI models will be made available to the scientific community online.

• DFG Programme: Research Grants