Episode 77: IR Trailblazers 1

Ezequiel Silva, MD, FSIR, speaks with host Sherief Ghozy, MD, on this special episode of the IR Trailblazers, a project from the SIR Resident, Fellow and Student (RFS) Section.

Listen to Dr. Silva discuss digital health, AI-applications and how interventional radiology had a responsibility to take ownership of this new era in health care. 

The following excerpt has been edited for flow and grammar.

Ezequiel Silva, MD, FSIR:

I think that the greatest opportunity and the greatest challenge for interventional radiology resides within digital medicine and to be more specific, resides within artificial intelligence. There's a futurist named Roy Amara, and he had an institute called literally the Institute for the Future which he founded in the 1960s. And one comment that he known for is called Amara’s Law, which is when we talk about technology, we tend to overestimate the short-term impacts, and we tend to underestimate the long-term impacts.

I wonder if we now, as a society and a house of medicine, and even as interventional radiologists, might be in that space where we're seeing a rapid change in digital health and a rapid growth of artificial intelligence, and where we're reacting to it in a way that might be overestimating the short-term impacts.

We need to be very mindful not to underestimate the long-term impacts and what they will do to clinical workflows, as well as how it will impact medicine in general and our specialty specifically.

Understanding digital health

Let’s do a deeper dive. Imagine digital health and artificial intelligence as an upside down triangle. The longer part is at the top and it narrows as we progress.

Digital health is at the top of that triangle. It is a very broad term. It's everything from telemedicine, telehealth, remote patient monitoring, remote therapeutic monitoring. It's even devices that can do tasks as seemingly simple but importantly meaningful as interfacing with patients and helping in their clinical decision making, even tracking the number of pills that they take every day from their medicine bottle to make sure that their therapeutic regimen is progressing appropriately.

So that's a very broad subject. Interventional radiology is a broad specialty, but we have a pretty narrow focus of what we do and the way that we take care of patients. When we think about all of the digital health tools that are available to us, we can almost do everything that a patient might require with a completely digital interface.

Patients don’t have to drive to a clinic or interface with a primary care physician for visits, screening or intake. We can do samples of the medical record. We can tabulate that data, we can do screening, we can do preoperative medications. Preoperative imaging interpretations mean everything that the patient might require for a complex intervention is done digitally. And a reason that might be relevant is it is quite conceivable that the first time a patient sees another human face to face is going to be an IR for the procedure. We are not at a point where technology can replicate what we as IRs can do.

From a procedural perspective, this shift is coming, but it's just not realistic as we sit here in 2026. And so I think this creates a very distinct opportunity for interventional radiology to not just be the first ones to bring all of this information together, but to be the first ones to translate digital information and guidance into a human-to-human interaction, and to do so in a way that's meaningful for that patient.

Taking responsibility for AI

Do you remember in November 2022, OpenAI sort of released itself on the universe? Chat GPT was one of the first large language models, foundational models, transformer models—you pick the term to describe it, but everything before that for the most part was what we would think of as “narrow AI.” It is sometimes called convolutional neural networks, where it is very focused on applications in the diagnostic radiology space. As in, looking at a CAT scan and determining the diameter of an abdominal aortic aneurysm. It reached a relatively narrow conclusion.

That simple. What we see now with these large language models is generation, or taking massive, massive amounts of data and not just reaching narrow conclusions. Not just two plus two equals four, but creating very far-reaching conclusions that will interrupt patient care in ways that are meaningful.

One way that is relevant for physicians collectively, and I think the introduction of biologists more specifically, is that the move towards ChatGPT and large language models creates a distinct opportunity and maybe even responsibility for us right now. ChatGPT was probably the fastest growing application in the history of mankind, from a technological perspective. There were 200 million users within the first two months, and that's fast on the internet. This technology commanded the medical space in a way that we as physicians probably have a responsibility to inform the next steps, to make sure it's accurate and to make sure the guardrails are in place.

I believe interventional radiology should take on the responsibility to be the first ones to interface with a patient who has thus far received most in their clinical guidance from a digital application. And not just to be the first, but to do so in a way that can translate into something meaningful from a human-to-human perspective. These interactions are informed not just by our experience, but also by our empathy and our ability to make decisions that are not just driven by an algorithm, but driven really by our humanity in a way that enables the best outcome possible for those patients.