Clinical AI is often evaluated as a product. Something hospitals pilot, measure, and either keep or abandon. Episode 5 of Radiology Rewired challenges that mindset. AI changes care only when it functions as hospital infrastructure, embedded into how decisions are made across teams and specialties.
In this episode, Dr. Vivek Singh speaks with Dr. David Stoffel, Chief Business Officer at RapidAI, about why many AI initiatives stall after early promise while others scale and endure. Drawing on experience across medicine, business, venture capital, and healthcare technology, Dr. Stoffel frames AI adoption as an organizational challenge rather than a technical one.
Many clinical AI tools demonstrate efficiency gains. Far fewer translate those gains into sustained clinical or financial impact at the health system level. As discussed in the episode, this gap fuels skepticism among hospital leaders.
According to Dr. Stoffel, the problem is rarely model performance. Most modern AI tools work as designed. The challenge is that insights often fail to influence downstream decisions. When AI output is visible to only one specialty, arrives too late, or sits outside established workflows, efficiency remains isolated. Hospitals see activity, but not impact.
Value emerges only when AI informs actions across care teams. That shift requires AI to move beyond point solutions and into shared clinical pathways.
Stroke care illustrates this clearly. AI did not transform stroke by elevating suspected cases alone. Its impact came from accelerating coordination and diffusing actionable, patient-specific insights. When radiologists, neurologists, emergency physicians, and interventional teams receive the same imaging insights at the same time, decisions happen earlier and outcomes improve. The technology matters, but the shared visibility matters more.
Because clinical AI crosses departmental boundaries, its effectiveness depends heavily on leadership. Without clear executive support at the adoption level, questions around access, ownership, and cost allocation can stall adoption.
Dr. Stoffel notes that hospitals seeing the greatest benefit treat AI as a strategic capability. They apply consistent evaluation criteria, align deployment with clinical priorities, and ensure AI outputs integrate into existing decision pathways. This approach reduces friction for clinicians and builds trust.
Restricting AI access to contain costs often undermines value. Infrastructure works because it is shared. When leaders design AI deployment around collaboration rather than silos, system-level benefits follow.
To explain how transformative technologies take hold, Dr. Stoffel draws on his experience with surgical robotics. Robotic surgery succeeded not by marginally improving workflows, but by enabling a fundamental shift in how procedures were performed. It moved care from open surgery to minimally invasive approaches, redefining standards of care.
Clinical AI follows the same adoption curve. Tools that add incremental insight on condition struggle to justify their place. Tools that expand access to timely, actionable information can reshape care pathways at scale.
The episode also explores AI’s shift from detection to prediction. Predictive models that analyze dozens or hundreds of clinical variables can identify risk patterns earlier than human observation alone.
Examples discussed include sepsis prediction tools that flag patient deterioration hours in advance, enabling earlier intervention and measurable reductions in mortality. These systems do not replace judgment. They support it by extending cognitive capacity and understanding.
However, predictive AI only delivers value when insights are presented in ways clinicians can interpret and verify. Poorly integrated predictions increase cognitive burden rather than reduce it. Governance, monitoring, and clinician feedback are essential to making these tools reliable and actionable.
A recurring challenge highlighted in Episode 5 is pilot-driven AI adoption. Many hospitals test individual tools in isolation, hoping to identify quick wins. Without a broader strategy, these efforts rarely scale.
Dr. Stoffel argues that successful organizations move toward platform thinking. Enterprise AI platforms allow multiple applications to share data, integrate into common radiology and care team workflows, and evolve under consistent governance. This shift mirrors how hospitals adopted imaging systems and EHRs as infrastructure rather than collections of point solutions.
AI is now at the same inflection point.
Radiology Rewired continues to explore how AI reshapes medicine by changing how care is coordinated across complex systems.
Episode 5 with Dr. David Stoffel offers a practical framework for leaders, radiologists, and clinicians navigating AI adoption at scale. The takeaway is simple. AI transforms healthcare only when hospitals design around it as infrastructure, not experimentation.
[▶ Spotify] [▶ Apple Podcasts] [▶ YouTube]
What does it mean to treat clinical AI as hospital infrastructure?
Treating clinical AI as hospital infrastructure means embedding it into shared clinical workflows and existing systems, governance, and decision-making across departments, rather than using it as a standalone tool for one specialty. When AI operates as infrastructure, its insights are accessible to the right teams at the right time and can influence system-level care decisions.
Why do many clinical AI tools struggle to show return on investment?
Most clinical AI tools perform well technically, but struggle to drive ROI when their insights do not change downstream actions. If AI output is siloed, poorly integrated, or disconnected from clinical workflows, efficiency gains remain isolated and do not translate into measurable clinical or financial impact.
How can hospitals move from AI pilots to scalable platforms?
Hospitals can scale clinical AI by shifting from pilot-driven adoption to platform-based strategies. This includes standardizing evaluation criteria, enabling shared data and access across specialties, integrating AI into existing workflows, and establishing clear governance so multiple applications can evolve together as part of a unified system.