The integration of artificial intelligence into radiology promises to reduce workload and enhance diagnostic efficiency. However, despite an abundance of commercially available AI solutions, implementation often falters due to misalignment with clinical workflows and institutional needs. To overcome these limitations, radiology departments increasingly engage in co-creation partnerships with AI startups. This collaborative approach allows both parties to merge their unique expertise, resulting in more clinically relevant, technically feasible applications. Yet, the process remains under-explored, particularly in terms of how such partnerships evolve and what factors determine their success. A recent multi-case study from a Dutch academic medical centre provides valuable insights by analysing three co-creation scenarios, ultimately offering a relational framework to guide these collaborations.
Modes of Engagement and Common Challenges
The study introduces a relational framework that categorises co-creation into three modes of engagement: resourcing, adapting and reconfiguring. Resourcing entails the allocation of existing resources—such as data or staff expertise—without altering underlying systems or strategies. This approach poses the least risk but may limit innovation. Adapting requires institutions and companies to extend or modify existing processes or structures to better accommodate the new technology. Reconfiguring goes further, involving a fundamental rethinking of workflows, systems and assumptions. While potentially disruptive, this mode also holds the greatest promise for transformation.
Each co-creation initiative must determine the appropriate mode of engagement, depending on institutional capacity, strategic priorities and the nature of the AI solution in development. However, challenges such as shifting startup priorities, limited technical resources, regulatory complexity and variable clinical engagement can hinder progress. In particular, differences in organisational culture and expectations between medical institutions and startups add layers of complexity. The study reveals that success depends on continuous alignment, proactive management and the ability to adapt strategies as circumstances evolve.
Case Examples and Key Insights
The study analysed three co-creation projects undertaken by the same academic hospital. The first case involved a musculoskeletal AI application for quantitative vertebral morphometry. The collaboration started with a balanced and adaptive approach but lost momentum when the startup redirected its focus due to limited resources. The hospital attempted to continue development internally, gaining experience in legal structuring and certification processes. Although the final product remained underdeveloped, the project yielded valuable lessons for future initiatives.
Must Read: Rethinking Nursing Innovation Through Co-creation
The second case focused on a chest X-ray solution. Here, the startup demonstrated high agility by rapidly pivoting based on clinical feedback. Initially intended for automated pathology detection, the tool was repurposed for normal/abnormal classification to help radiologists prioritise complex cases. The project achieved clinical implementation and integration into the hospital's imaging systems. By involving radiologists throughout the process and refining the tool based on real-world feedback, the partnership generated tangible efficiency gains.
The third case addressed chest CT imaging and illustrated the benefits of sustained, balanced engagement. Initially conceptualised as a search tool, the application evolved into a comprehensive analysis platform. Both the hospital and the startup invested heavily in adapting workflows, integrating systems and responding to user feedback. This deep level of collaboration led to successful clinical integration, improved diagnostic processes and new research opportunities.
Across all cases, success was tied not only to technological outcomes but also to the quality of the collaboration. Factors such as early clinician involvement, shared project leadership and iterative development cycles were found to be crucial for alignment and impact.
Strategic Applications of the Framework
The relational framework proposed by the study provides a practical tool for mapping and monitoring co-creation engagements. It enables institutions to assess their position along the resourcing-adapting-reconfiguring spectrum and adjust their strategies accordingly. For example, if a project becomes unbalanced—favouring one party’s engagement over the other—realignment actions can be taken to restore mutual commitment.
The framework also supports strategic partner selection. By choosing collaborators with aligned clinical interests or subspecialty expertise, institutions can build stronger foundations for co-development. While the study did not mandate strict selection criteria, the hospital in question approached each partnership on a case-by-case basis, considering institutional goals and available resources. Pilot projects were used to test compatibility before scaling up collaboration.
Another benefit of the framework is its ability to inform both proactive and reactive decision-making. Proactive steps include setting milestones, formalising project roles and establishing shared objectives. Reactive measures may involve revisiting resource allocations or modifying project scope in response to unexpected challenges. By combining structured project management with adaptive engagement, institutions can better manage the uncertainty inherent in developing and implementing AI in clinical environments.
Co-creating AI solutions in radiology is a dynamic, multifaceted process that demands ongoing collaboration, flexibility and strategic oversight. The study from the Dutch academic centre highlights the importance of understanding engagement modes, aligning stakeholder interests and learning from experience. By using the relational framework to guide and assess their partnerships, medical institutions and startups can enhance the likelihood of developing AI tools that are not only technically sound but also clinically valuable. Ultimately, the success of AI in radiology will depend as much on the quality of co-creation as on the capabilities of the technology itself.
Source: European Radiology
Image Credit: iStock
