DECADE is divided into five subprojects. Different steps of the work packages are carried out at our partner sites in Dresden, Bonn, Heidelberg, Mainz and Düsseldorf.
Work Plan
Build a multi-institution Swarm Learning (SL) network in Germany for biomarker discovery in colorectal cancer and set up the technical infrastructure at each site.
- Develop the pipeline and tailor it to the specific needs of biomarker discovery in colorectal cancer (CRC).
- Deploy pipeline: set-up at participating institutions and empower researchers to use it.
- Use the SL pipeline to provide medical image data to the consortium and complement the datasets at the partner sites without sharing actual data.
Understand hereditary cancer predisposition: Prediction of pathogenetic mechanism, molecular subtypes and outcome from histology images of precancerous lesions
- Identify Lynch Syndrome (LS) patients based on adenoma or macroscopically normal tissue.
- Perform AI-based risk stratification in Lynch syndrome patients and predict prognosis.
- Detect distinct histopathological features using generative models to understand underlying immunological processes. AI-based analysis of histology specimens yields spatially resolved sensitivity maps, allowing to identify tissue regions or cell types associated with molecular alterations or clinical outcomes. Next, these “highly predictive regions” (HPR) will be linked to spatial transcriptomics and proteomics data.
Predict underlying pathogenetic mechanism, molecular subtypes and clinical outcome of colorectal cancers (CRCs) from routine pathology slides and identify histopathological patterns characteristic of certain molecular subtypes of Lynch syndrome
- Distinguish between hereditary and sporadic CRCs based on distinct biological features such as immunological characteristics.
- Dissect molecular CRC subtypes within Lynch syndrome for improved early diagnostics and disease prevention.
- Identify features using generative models (xAI).
Progressed primary CRCs: Predict pathogenic mechanisms, molecular subtypes and clinical outcome and improve differential diagnosis of liver metastases
- Generate a comprehensive clinical cohort.
- Adjust the preprocessing pipeline and set up the environment at all sites.
- Perform additional molecular and immunohistochemical analyses.
- Train, validate and test the models.
- Analyze data and apply explainable AI (xAI) to increase model transparency and to potentially discover new morphologic features associated with prognosis, molecular aberrations or therapy response.
Validate findings, manage outreach and patient involvement and expand the swarm consortium
- Validate swarm results by blinded external analyses to demonstrate the robustness of the respective biomarkers.
- Coordinate outreach and patient involvement.
- Expand the consortium to establish a German-wide SL-AI network in translational cancer research.