The increasing use of AI in patient diagnosis raises questions about workflow and governance. Twelve U.S. health systems have formed a Diagnostic AI Consortium with vendor Aidoc to develop shared standards for how diagnostic AI is evaluated and governed. Elad Walach, CEO and co-founder of Aidoc, spoke with Healthcare Innovation about the consortium’s goals and health systems’ urgency to address issues around capacity as well as diagnostic errors and delays. 



HCI: Is there value in the size of this group? Do you need the scale or the amount of data that these health systems represent to improve the quality of what you're able to do with algorithms?

Walach: It’s not about the algorithm; it’s much more about the validation and diversity of workflows. It's not really about the size. The size is a side benefit of the diversity, because what we want to recommend are workflow changes and playbooks. AI has some diagnostic capabilities, but if you want to implement that in a way that really improves physician workflow and improves patient care, how do you do that? That's really what we're aiming for. Of course, everything we do is really about augmenting physicians, not replacing them.

HCI: Aidoc has solutions in radiology, cardiology neuro and vascular. Is this focusing in on radiology or does it involve several areas at once?

Walach: The whole point is to be as as broad as we can. This is going to be a multi-year journey. On top of the areas you said, we also want to focus on emergency department workflow, which is in very dire straits these days, and population health. So essentially it is across service lines, and the principal question is: If AI can actually create a first draft of information in a file, what does it mean for workflow, and how can you reconfigure and redesign workflows from the get-go?

HCI: What was involved in convincing these health systems to come together to do this work? 

Walach: First, I think the key in compelling them was understanding there will be shared learnings with other systems. Second, it's an ecosystem play. It's really not about Aidoc to some extent. Yes, we're facilitating this, but if other AI companies join, that's fine, and we'll share. The idea is really not about us. It’s a diagnostic AI consortium; it’s not the Aidoc consortium, right?

HCI: Do you picture a couple of leaders from each organization coming to meetings, reaching agreement on something, then going back to their organizations to work on it, and then reporting back?

Walach: We are going to have these governance meetings of the full consortium, and we’re each going to work on projects that are specifically well-suited for that system. We're going to implement, work, come back, share learnings, iterate, and do it again. That is how we imagine it.

HCI: What will likely be the next thing the public hears from the consortium? Are there going to be reports or updates about the work?

Walach: We are planning a very big data milestone within a year — to publish the first findings, but we hope to do it much earlier than that.