Demetri Giannikopoulos

To Buy or Not to Buy? The Guiding Principles of ECLAIR When Choosing an AI Vendor

Artificial intelligence (AI) is becoming the standard of care, yet questions still linger when it comes to codified governance. Beginning as point solutions intended to solve singular hospital pain points, clinical AI has since widened its scope and is making a crater-sized impact throughout the healthcare enterprise. Yet with a growing body of clinical validation, there is still no single entity holding full control of AI oversight. Various regulating bodies, credentialing and member organizations, however, are releasing documents to offer highly sought after guidance for those starting their AI journeys. Enter “Evaluating Commercial AI Solutions in Radiology,” better known as ECLAIR.

ECLAIR acknowledges the crucial role radiology plays in adopting new technologies (we talk about that more here), providing a list of 10 key questions radiology departments ought to ask when weighing out their options for AI vendors. Below is a set of brief takeaways from each of the questions posed in the ECLAIR guidelines.

  1. Defining Clinical Need: Clarify the target users and the specific problem you want AI to solve. How will you measure the success of the AI solution post implementation? Will it be a diagnostic aid, triage tool, or something else entirely?
  2. Balancing Benefits and Risks: Evaluate the potential benefits and risks of the solution. Consider its impact on clinical outcomes, workflow efficiencies and any associated risks.
  3. Validation and Performance: Understand the rigorousness and independence of the algorithm and its validation process. Ensure it has been thoroughly assessed for performance across relevant patient populations and imaging modalities. 
  4. Integration and Workflow: Assess how seamlessly the AI application can integrate into existing workflows and whether it is interoperable with current software systems.
  5. IT Infrastructure Requirements: Consider the IT infrastructure demands and engage with your IT team early and often to address any potential hurdles. (If you’re in IT, here’s a separate list of questions worth asking an AI vendor.)
  6. Regulatory Compliance: Ensure the solution complies with medical device and data protection regulations in the target country, understanding the implications for implementation. For instance, adoption in US systems require FDA clearance while EU systems require CE certification.
  7. Return on Investment (ROI): Conduct an ROI analysis to better understand the economic viability of the AI solution in question. Ensure it works within your facility’s budgetary constraints. Click here for a deep dive on the ROI of clinical AI.
  8. Maintenance and Support: Discuss ongoing maintenance and support, including future proofing against the inevitable evolution of your IT department, systems and requirements.
  9. User Training and Support: Evaluate the availability of user training and ongoing support mechanisms to facilitate effective adoption and utilization of the AI solution by your staff. As Debi Taylor, RN, wisely observed, “Intuitive functionality does not equal intuitive adoption.”
  10. Error Management and Surveillance: Establish protocols for managing potential malfunctions or erroneous results (false positives and false negatives, for example), emphasizing the importance of post-implementation surveillance and protocols to ensure constant improvement.

Want to learn more about other AI regulations and guidelines? Click here for an ever evolving resource of expert-driven, streamlined takeaways.

Explore the Latest AI Insights, Trends and Research

Demetri Giannikopoulos