The Artificial Intelligence for Health Imaging (AI4HI) cluster has published the position paper ‘Data infrastructures for AI in medical imaging: a report on the experiences of five EU projects’ in the European Radiology Experimental journal.

The manuscript summarizes different key points regarding the architecture, data models, General Data Protection Regulation (GDPR) considerations, and curation processes adopted by the five EU projects that made up the cluster: INCISIVE, Chaimeleon, EuCanImage, ProCancer-I, Primage. All these projects are working collaboratively towards developing big data infrastructures based on European, ethical and GDPR compliant, quality-controlled, cancer-related, medical imaging, and related patient’s data platforms, in which both large-scale data and AI algorithms will co-exist.

After describing the different approaches and solutions that the projects are currently deploying, the article focuses on their common challenges and gives recommendations based on their experiences on the following aspects: architecture design (centralized/federated/hybrid), data models and standards used (DICOM, OMOP-CDM/SNOMED-CT/HL7 FHIR, etc.), use of cloud-agnostic or cloud-dependent solutions, considerations on GDPR, consent versus patient information, accountability, data minimization, deidentification and curation processes.

Finally, the document mentions the cluster’s initiative to launch the EuCAIM (European Federation for Cancer Images) project to further “harmonize practices, contribute to standards, ensure interoperability and in-fine, raise trust in data sharing, and advance in AI development in cancer imaging”.

The article is available in open access on this link.