About the project
The RACOON-LCS project aims to establish a reliable and quality-assured infrastructure to support national lung cancer screening (LCS) in Germany. Based on the existing RACOON infrastructure, RACOON-LCS will create the first nationwide data room for screening data.
This promotes cooperation between sites and enables long-term standardisation, quality assurance and scientific analysis of the screening processes. A central element of the project is the integration of AI-supported detection systems (CAD), which help to detect and classify round lung lesions at an early stage, analyse the lung structure and better assess the risk of malignant changes.
The most important facts at a glance
The overarching goal of the RACOON-LCS project is to utilise and further develop the publicly funded RACOON Research Infrastructure for university support of lung cancer screening in Germany and to establish a "German Data Space" for lung cancer screening data. At least two AI-based, computer-aided diagnosis systems (CAD) are to be integrated into the existing infrastructure and initially analysed qualitatively using retrospective image data analysis. From mid-2026, screenings will be carried out at the partner sites of the RACOON infrastructure and the pseudonymised image data will be uploaded to the central RACOON-CENTRAL instance and analysed there. The focus is on ensuring the quality and consistency of the image data sets across all locations. The project is investigating the influence of AI on the accuracy and efficiency of detecting lung nodules, analysing lung structure and making medical diagnoses in accordance with national guidelines.
In the first part of the study, existing CT images of the lungs and associated findings from the Heidelberg and Munich sites will be analysed. The aim is to prepare the CAD systems used for future analyses as part of lung cancer screening. Among other things, it will be investigated whether inflammatory changes, which can occur depending on the season, influence the assessment of lung round foci by Artificial Intelligence, especially in the case of rather unclear changes in the lung tissue. Body composition analyses will also be carried out. The aim is to test whether certain body characteristics are suitable as possible indications of an increased risk or as biomarkers in the context of preventive measures. Another aim is to test whether modern language models (AI systems that can formulate texts in an understandable way) can process medical findings from the CAD system in such a way that they are easier to understand for patients and laypeople with medical training.
In the prospective part of the study, new CT data (low-dose CT, LDCT) from lung cancer screening with low radiation dose as well as regular CT images in which lung nodules were discovered by chance are analysed. This data comes from several participating centres in the NUM. The aim is to test how reliable and transferable the computer-aided evaluation systems (CAD) work at different locations. The study will also investigate whether differences in the performance and image processing of the CT examinations influence the performance of the Artificial Intelligence. In addition, analogous to the first part of the study, it will be tested whether large language models (AI text systems) can convert structured medical findings into comprehensible, coherent texts for different target groups.
The greatest challenges of RACOON-LCS arise from the early stage of development of lung cancer screening in Germany. At the start of the project, it is not yet possible to predict how high the participation rate in screening will be and over what period of time the prospective data acquisition can take place. Therefore, an initial retrospective analysis of existing CT data sets will be used to test the planned workflow at an early stage and at the same time to address relevant scientific questions.
For the prospective part of the project, the onboarding of all 35 university partner sites and the fulfilment of ethical and data protection requirements pose key challenges. However, experience from previous projects shows that cross-location communication can be efficiently supported via the established Confluence research portal. In addition, a kick-off meeting and regular consultation hours should facilitate the integration of the partner locations and ensure a continuous exchange.
Another challenge is the harmonisation of image data across locations. As CT imaging and reconstruction parameters can vary between the participating centres, differences in image quality, protocols and data formats must be systematically recorded and taken into account. This is particularly relevant as such technical differences can influence the performance of CAD systems.
The project is being carried out in five work packages:
WP1: Project coordination and management
WP2: Local data collection and annotation
WP3: Quality control and AI-supported analysis
WP4: Statistical analysis
WP5: Integration of large language models
The overall management of the project is carried out by the local study coordinator and the site IT specialist at Heidelberg University Hospital. They coordinate the organisational, technical and technological processes and support the integration of the participating partner sites.