Cancer imaging data key to diagnostic AI tools
With there being thousands – perhaps hundreds of thousands – of typeArtificial intelligence (AI) tools are increasingly being used to help diagnose medical conditions. To do this effectively though, large amounts of data are needed to properly train and test these tools. The EuCanImage(opens in new window) project sought to build a pan-European repository of cancer imaging data to achieve this aim. The idea was to combine this resource with clinical information, including lab results and tumour analyses, to help build up a fuller picture of a patient’s profile. Another key element was ensuring diversity within the dataset, to enable AI tools to take account of different populations and European countries. Developing a pan-European resource is also critical for low-incidence diseases such as liver cancer, where small amounts of data tend to be contained in local repositories.
Medical imaging and clinical data
Existing networks and data infrastructures in Europe were brought together to build the platform. These included the BBMRI-ERIC based in Austria, which gathers biological and tissue data, Euro-BioImaging, which offers open access to biomedical imaging, and the European Genome-phenome Archive, which contains genetic, phenotypic and clinical data generated for biomedical research projects. “We also learned from the Cancer Imaging Archive in the US, so as to avoid reinventing the wheel,” notes project coordinator Karim Lekadir from the University of Barcelona(opens in new window). “Other research organisations, data scientists and clinical sites were also involved.” The consortium also included companies to build the AI tools along with associations such as the European Association for Cancer Research. “Some 20 institutions in total from all over Europe were involved,” adds Lekadir.
Addressing unmet clinical needs
The platform succeeded in bringing together 25 000 new cancer cases across Europe, with a focus on breast, colorectal and liver cancer. Real case studies were used to help highlight unmet clinical needs. Finding small tumours in the liver for example was highlighted as something that AI could help with. “This platform is about solving medical problems, so we first talked with clinicians and took it from there,” says Lekadir. Another key element of the platform is that it is built on federated learning. “This means that data can stay where it is,” notes Lekadir. “Instead of hospital data moving towards researchers and developers, tools go to where the data is. This has become a reference solution in Europe. Cancer Image Europe(opens in new window), a flagship EU programme, is now linking repositories like ours together, and the initiative is using our federated learning platform to do this.”
Deploying AI health tools
In addition to the data platform(opens in new window), the project team built on previous work on AI and ethics to produce the FUTURE-AI Guideline(opens in new window). Other projects were invited to participate in the development of this resource. “This is important as without trust, this is not going to work,” remarks Lekadir. “A paper was published in February 2025, which has since been cited more than 300 times.” This aspect of the project is being continued through the COMPASS-AI(opens in new window) initiative, which aims to promote the responsible and effective integration of AI into clinical settings. The project has also provided lessons on legal governance to enable data sharing. Next steps include integration of the platform within Cancer Image Europe, which will ensure that the project’s successes are linked with others and built upon. “We are moving from AI developed in labs to creating AI solutions from real-world hospital data,” says Lekadir.
