Teleradiology: High quality versus high volume

By Dr. Josh Ewell, Founder & President, Frontier Radiology

About a decade ago, at RSNA, I met with a physician from a prominent Manhattan health system who told me the story of how his hospital recovered from Hurricane Sandy. The storm surge flooded the basement where much of their MR imaging was housed, and the department lost several of its scanners in a single night. However, the patients did not go anywhere, and the community still needed the same imaging, so the department was left with a decision that sounds simple on paper but is anything but in practice: scale back to what its remaining equipment could handle, or find a way to do more with less.

They chose the latter. They pressed research scanners into clinical service, brought in mobile units, extended their hours into nights and weekends, tightened their protocols, and rebuilt their workflow around the people they still had rather than the scanners they had lost. As he told it, they came out of it doing more imaging than before, on fewer scanners.

I have thought about that conversation many times since, mostly because of what it says about how resourceful hospitals are. When patients need imaging, health systems find a way to provide it, whether through better processes, better scanner technology, smarter scheduling, or simply more scanners, and the capacity to image patients has grown steadily for as long as I have been in medicine.

Yet, every one of those studies still ends up in front of a radiologist, and interpretation is the one step in the chain that has not scaled the same way. The radiologist remains the bottleneck, and most of us feel it every shift.

AI is part of the answer to that bottleneck, and we use it ourselves. AI that helps identify findings, assembles the history, and removes friction from the shift gives radiologists real time back, and I think it belongs in every reading room.

But much of the conversation around AI in radiology is about replacing the radiologist’s report altogether, with the radiologist left to review AI output at whatever pace it sets. That is the reverse centaur I described in “The two black boxes and the centaur,” where the radiologist becomes a rubber stamp, and I think it would be a mistake for patients, for radiologists, and for the hospitals themselves.

Our philosophy is different. We use the same tools to raise the quality of the work and remove the friction around it, because a role measured only by the reports we produce treats us as if our jobs were confined to interpreting the images, when that has never been the case.

Our consultation often begins before the patient is imaged, as we help guide the right study, and it does not end with the signature on our report. But because our report is the most visible component of our work, and the easiest to track, many people who evaluate radiology as a service line forget how much we contribute to the process leading up to the imaging, not to mention after the report is completed.

A CT tech messaged me recently to ask how to protocol a CTA, which is about as ordinary an interruption as there is in this job. There was a fast answer available, and I have given it many times before: follow the standing protocol, get the patient on the table, and move on to the next study. Instead, I took a couple of minutes to discuss the patient with the technologist and look at the chart.

The patient was septic, and the clinical question was whether a portal vein thrombus was the source, which was a reasonable question to ask. However, the patient had already had two CTs of the abdomen and pelvis that week; I had read one myself three days earlier, and a colleague had read the other earlier on the day of the call. The portal vein thrombus was stable from three days earlier and unchanged from months before, there was no inflammatory change around the vein, and nothing suggested pylephlebitis. The failure to find a source of sepsis on two CTs that week was the reason for the new order; our reports had not given the physician the answer she was looking for, so she ordered another exam. I asked the tech to hold the study and have the ordering physician call me.

The physician was frustrated and hesitant to even call me, and at first she insisted through the technologist that I simply do the imaging, because she was the one seeing the patient and had better clinical insight than I did. She was right, and that was exactly why I wanted to talk with her.

What she wanted was a study that would answer her question, but what I had to tell her was that no study would. The imaging the patient already had answered her question about the portal vein thrombus; the source of the sepsis was somewhere else, and another scan was not going to find it. By the end of the call she was grateful that I had taken the time to talk it through with her. She was still understandably frustrated that she was no closer to an answer about why her patient was septic, but she could see that the patient had already been imaged extensively, and that one more study would not give her the information she wanted.

For that patient, my value was not in reading another scan; it was in saving an unnecessary one. It saved the patient the radiation and the cost of another study, it saved the hospital the wasted resources, and it saved us from reading imaging that would not have added benefit to the patient’s care. Perhaps most importantly, it spared the hospital the opportunity cost of that scan by freeing the table for another patient who did need imaging, which is its own kind of newfound capacity and one that no workflow redesign can create.

That is a decision radiologists are better positioned to guide than anyone else in the chain. We are the ones who can see what imaging has already shown, judge whether more imaging can show anything else, and know which study, if any, actually answers the question being asked. AI can already suggest a protocol and provide guidance, and it will only get better at it, but deciding whether a septic patient needs a third CT in a week is a judgment that only a radiologist should make.

Still, we can only make it if we have the time to talk and the referring team can reach us when it matters, and that is exactly what disappears when the answer to the bottleneck is to add volume without removing friction, which takes time from the radiologist instead of creating it.