Access to high-quality medical data is essential for developing reliable healthcare AI, but navigating privacy requirements, institutional procedures and technical barriers remains a major challenge for innovators. Croatia’s European Digital Innovation Hub AI4Health.Cro has developed a structured approach that brings together healthcare institutions, researchers, technology developers and public authorities. Through secure data processing, expert support and three annual innovation challenges, the project demonstrates how responsible access to health data can support the development and testing of AI solutions addressing real clinical needs.
Artificial intelligence holds significant potential for healthcare, from supporting diagnostic decisions to identifying patients who may need additional attention. But developing useful AI tools requires more than technical expertise.
Innovators need access to relevant, well-prepared medical data, an understanding of clinical challenges and the ability to work within strict privacy and regulatory requirements.
For start-ups and SMEs, these conditions can be particularly difficult to meet. Hospitals and public authorities, meanwhile, must ensure that sensitive patient information is handled responsibly and that any secondary use of data is appropriately governed.
Croatia’s European Digital Innovation Hub AI4Health.Cro addressed these challenges by developing a structured approach to connecting healthcare needs, secure data access and AI innovation.
Rather than treating data access as an isolated technical problem, the project brought together the institutions and expertise needed to support the development of healthcare AI under controlled conditions.
At the beginning of the project, AI4Health.Cro established a dedicated data group involving 10 of its 16 consortium partners.
The group combined expertise in data acquisition, management, anonymisation, modelling, clinical practice and regulation. Its work contributed to the development of a regulatory sandbox model, referred to as a Secure Processing Environment, designed to support the extraction, anonymisation, preparation and controlled processing of real-world medical data.
The approach drew on procedures developed within the European Health Data Space framework, with key involvement from the Croatian Institute of Public Health, the Croatian Health Insurance Fund and the Ministry of Health. The process connected several essential steps: defining a healthcare problem, identifying the necessary data, establishing appropriate access conditions, preparing and anonymising datasets, specifying rules for secondary use and providing computing resources and expert support.
The Ruđer Bošković Institute also compiled a catalogue of publicly available datasets suitable for AI development, including information about access conditions for different user groups.
Together, these activities created a more structured starting point for innovators seeking to develop and test healthcare AI solutions.
AI4Health.Cro applied this approach through three annual innovation challenges, each centred on a specific healthcare problem.
In 2024, participants explored how AI could help predict hospital readmissions within 30 days of discharge. The competition attracted 99 applicants in 27 teams. Participants developed predictive models and prototype interfaces intended to make individual predictions understandable.
In 2025, the focus shifted to breast cancer diagnostics. A total of 106 competitors applied through 34 teams, with 14 teams selected for further development. The challenge involved predicting BI-RADS assessment categories, identifying suspicious lesions in mammography images and developing prototypes for radiological decision support.
In 2026, the challenge addressed medication adherence in type 2 diabetes. It attracted 150 participants in 44 teams, with 24 teams comprising 85 competitors advancing to the development stage. Supported by AstraZeneca and Novo Nordisk Hrvatska, the competition explored how predictive models could identify patients at increased risk of interrupting prescribed treatment.
The winning team, Insulogic, developed a prototype platform combining risk identification with a mechanism intended to support clinical follow-up.
Across all three editions, the challenges connected AI development with practical healthcare questions while giving participants access to structured tasks, relevant data or data-based problem settings, mentoring and expert feedback.
The project highlighted a key lesson: access to data alone does not create useful healthcare AI.
Innovators also need clearly defined clinical problems, appropriate infrastructure, support in interpreting results and a pathway towards clinical validation.
For start-ups and SMEs, this approach offers opportunities to test ideas before committing further development resources. For clinicians, it creates a way to contribute medical expertise and help assess whether proposed solutions respond to real healthcare needs.
For researchers, students and developers, the innovation challenges provide opportunities to work in multidisciplinary teams and explore applications of AI in healthcare.
The experience also informed AI4Health.Cro’s wider Test Before Invest services, including support for secure data access, model interpretation, clinical feedback and expert mentoring. Importantly, the innovation challenges produced prototypes and development experience, rather than independently validated evidence of improved patient outcomes.
The AI4Health.Cro experience offers a practical example for European Digital Innovation Hubs and other organisations supporting innovation in regulated sectors.
Its distinctive feature is not a single technology or dataset, but the combination of institutional collaboration, controlled data processing, healthcare expertise and challenge-based experimentation.
By bringing these elements together, the project established a repeatable approach for moving from a defined healthcare problem towards an AI solution concept that can undergo further testing and validation.
As European healthcare systems explore how to make better use of health data, this experience offers a transferable lesson: responsible AI innovation depends on building the right conditions for data, technology and clinical expertise to work together.