AI That Works for Radiologists: Dr. Andrew Del Gaizo’s Perspectives

If radiology’s future with AI is going to work, it must work for radiologists, not the other way around.

In this Q&A, Andrew Del Gaizo, MD, Chief Medical Information Officer at Rad AI, shares how he distinguishes between diagnostic and operational AI, why trust and explainability are central to safe adoption, and what true success could look like five years from now. 

You’ve talked about “operational AI” as being distinct from “diagnostic AI.” How do you sum up that distinction, and what does it actually look like for radiologists on a day-to-day basis?  

Diagnostic AI helps the radiologist decide what’s in the pixels. For example, intracranial hemorrhage (ICH) detection or pulmonary embolism (PE) triage. Operational AI helps us decide what happens around the pixels, like case assignment, impression drafting, follow-up orchestration, quality nudges, workload leveling, payment denials prevention, etc. 

In practice, they both have potential and can be used together to reduce friction, but not all AI is the same. It’s essential to understand the needs of your practice and evaluate the real utility of a tool in your specific patient population.

Given your recent study on automation bias, how do you see subspecialization influencing radiologists’ trust in AI, and what strategies are effective in preserving independent clinical judgment?

In our survey work, we found that non-neuroradiologists were more likely to lean on the ICH model’s output than neuroradiologists. Subspecialization brings more experience and skepticism that buffers the user from over-reliance. Some effective strategies to mitigate this include:

• Certainty outputs: AI that shows calibrated confidence and explains the “why,” rather than offering a binary yes/no output, 
• User education: Training on the common AI failures
• Guardrails: Routing low-confidence cases to subspecialists, so human expertise leads when the model is uncertain.

Let’s fast forward five years. What do you hope radiologists will say AI finally got right – something that makes them wonder how they ever worked without it?

My hope is that AI exponentially delivers on its efficiency promise, giving time back to the doctor to look at the image rather than perform clerical duties. 

Further, I hope AI brings clarity to the decisions we make (less hedging) and peace of mind that downstream care will be appropriate (no patient lost to follow-up). I think a big step toward this will require explainable AI, where AI flags its uncertainty and knows when to step aside.

Finally, I see a world where the joy of providing patient care returns. If we spend less time hunting for data and performing administrative tasks, there is tremendous potential to reduce the burnout symptoms many doctors experience. 

Personally, this is why I leaned into the CMIO/entrepreneurial path: to create tools that give radiologists their time and their voice back. When we do that, the altruism that drew many of us to medicine can shine through.

Continue the Conversation

Dr. Del Gaizo’s perspective adds another dimension to our CMIO Q&A series, where radiology leaders reflect on how AI can meaningfully transform healthcare.

To see how these ideas build across the series, revisit our previous interview with Dr. Rishi Seth, who discusses how radiologists themselves hold the key to turning AI from hype into real-world impact: Dr. Rishi Seth on Why Frontline Radiologists Hold the Key to AI Success.