Rethink Imaging

EP 54 • October 1, 2026

Why the Average Patient Cannot Exist in Medical Imaging

Featured Guest

Dr. Ehsan Samei, PhD, DABR, FAAPM, FSPIE, FAIMBE, FIOMP, FACR

Chief Imaging Physicist • Duke University Health System

Listen Now31 min

Also on:

Let's Talk

Beyond dose

Ask us what your imaging data can actually show you.

Start a Conversation

Share This Episode

Subscribe to Rethink Imaging

Medical imaging teams often begin with manufacturer-recommended protocols and broad dose benchmarks. Those tools are useful starting points, but they cannot account for every scanner configuration, clinical question, or patient circumstance. When protocols are treated as fixed rules, imaging can become centered on the machine or an abstract average rather than the person being examined.

On this episode of Rethink Imaging, host Chris St. John speaks with Dr. Ehsan Samei about the clinical role of medical physics and the shift from “physics in medicine” to “physics for medicine.” Dr. Samei explains why radiologists, technologists, and physicists need one another to manage the many variables that shape image acquisition. He also distinguishes harmful protocol variation from appropriate personalization, explores how patient context can be organized through EMR tags and smaller data groupings, and argues for professional guidance that balances dose, image quality, and diagnostic utility.

Host

Chris St. John

Host, Rethink Imaging • Imalogix

Featured Guest

Dr. Ehsan Samei, PhD, DABR, FAAPM, FSPIE, FAIMBE, FIOMP, FACR

Chief Imaging Physicist • Duke University Health System

Watch the Episode

Key Takeaways

Medical physics is most valuable when it applies quantitative and analytical expertise directly to clinical needs, moving from “physics in medicine” toward “physics for medicine.”
Imaging systems contain many adjustable variables, including dose, reconstruction, field of view, and pitch, so vendor-recommended protocols should be treated as a starting point rather than a universal endpoint.
Patient-centered imaging depends on collaboration among radiologists, technologists, and medical physicists because each group contributes distinct expertise to interpretation, patient interaction, and image acquisition.
Clinical teams should reduce bad variability across similar cases while preserving good variability that adapts protocols to a patient’s anatomy, history, preferences, and clinical context.
Dose thresholds and diagnostic reference levels are most useful as guidance within a multidimensional review of image quality, diagnostic utility, and patient needs, not as simple pass-or-fail limits.

Full Transcript

Rethink Imaging Podcast Transcript

Guest:

Host: Chris St. John

CHRIS ST. JOHN 00:01:42 Welcome back to Rethink Imaging. I am so thrilled to have probably the closest friend of the podcast that we have, Dr. Ehsan Samei, podcast regular, medical physics legend, and somebody who I think I can call a friend at this point. Dr. Samei, welcome back again.

DR. EHSAN SAMEI 00:01:43 Thank you, Chris.

CHRIS ST. JOHN 00:01:44 I am super excited to have you back because we’ve been doing this show for two years now, which is crazy to me. I believe you were our second guest ever. And since then, we’ve gone on a lot of different journeys. We’ve talked way too much about those. We’ve talked a significant amount about AI and imaging. We’ve been all over the place talking about technologies and people and roles and all of this different stuff. But I want to get back to basics a little bit and get back to the core of medical physics, because I feel like we’ve gotten so inside baseball and we’ve gotten so granular with some of these conversations that we’ve been drifting from the core competency of medical physics’ role in imaging. I feel like you are the perfect person to have that conversation with me. Super high level: when you think about medical physics’ role in a clinical healthcare setting, what is the role of the physicist? What is the role of physics? And where does it fit in this organizational map, at least today?

DR. EHSAN SAMEI 00:03:08 I guess the answer lies in the very term itself. Medical physics is physics and medicine meeting one another. In the context of radiologic medicine, which is the primary focus that we have been dealing with here, radiology and physics meet one another. Many of us think of medical physics as physics of medicine or physics in medicine, and that is definitely an appropriate depiction of what medical physics is—it’s the physics related to medicine, or radiology in this case, and physics as it’s practiced in medicine. But increasingly, we have been moving towards what I call “physics for medicine.” In other words, medicine is the ultimate calling that we have in this space. The question that I ask myself, and physicists in the field should ask themselves, is: what is it that medicine needs from physics to make it better?

One way to think about it is that physicists, by their training and competence, are quantitative scientists. We spend our time going to graduate school, getting doctorates and master’s degrees and so on, trying to understand data, trying to interpret data. We plot data. We look at trends. We look at predictions. We have principles. We do experiments. We look at the results that we get from experiments—quantitative results from experiments—to see how those results verify or invalidate the assumptions that we had about the area that we practice. This is a wiring that medical physicists have. In the modern practice of medicine, I think that wiring can be useful. It gives a quantitative edge and analytical approach towards understanding medical processes. Science is very quantitative in nature, so we are individuals who add quantification, an analytical mindset, and methodological approaches towards the practice of medicine.

CHRIS ST. JOHN 00:05:42 One of the things that I feel like I have noticed in my last two years working in imaging tech is—I hear from a lot of physicists, and I talk to a lot of physicists—very often, there can be a sense that radiologists and physicists, the channels of communication between them are not always open and flowing in the ways that maybe the system was designed to or that we would hope. In fact, I’d rather not say who it is, but I recently spoke to a physicist, [unclear: sounds like “Hassan”], who I think you know pretty well, who recently changed where they’re working. One of the key things that they said about their new role is, “Man, the radiologists are just so involved, and they really, really care. They’re coming to me about the images, and we’re talking through things, and we’re working through things at a level that I haven’t seen before,” which to me sounds beautiful and like an ideal, but it is absolutely not the norm. I’m curious from your perspective, why do you think that that perhaps isn’t quite so normalized in practice?

DR. EHSAN SAMEI 00:07:14 Multiple reasons. One of the hallmarks of a civilized workforce or society is that we use experts for different areas of need. I never ask an electrician to do the plumbing in my house, or vice versa. Radiologists have very specialized skills in terms of understanding images and putting the images that they’re getting in the context of broader patient care. They look at the patient chart; they have a lot of information and deep insight into signatures of disease and what is normal versus abnormal. That’s a space that they can fully claim as theirs.

For us as physicists, our job is to understand how anatomical and physiological properties of the body can be best visualized, and how we can extract that information in the most effective, efficient, cost-effective, and dose-effective fashion. That is our wiring. This is our space. So the expertise of radiologists and physicists is very complementary. In an ideal world, they work with one another. We both work in the context of medical imaging, but radiologists are leaning more toward interpretation, while we are leaning more toward the technological aspects of image acquisition and image formation for the care of the individual patient. That’s how we chart the territory of radiological practice.

Of course, we are not the only two stakeholders involved. There are technologists who engage with the patients and image the patients. If we respect each other’s expertise and work together, we can synergize our expertise towards doing right by the patient, which is ultimately the goal of medicine anyway. We need to work together to deliver the best care for the patient.

Now, why is it that in many places it doesn’t work very well? One reason is that we commodify the technology. We feel like, “Why do you actually need to have a physicist? Can’t I just go to the vendor, buy the machine, bring the patient to the room, ask the technologist to acquire the images, and then read the cases?” What’s wrong with that model? Many places perhaps use that model. The problem with that is we forget that the imaging process has—using physics jargon here—too many degrees of freedom. Another way of saying it is there are too many things you can adjust, and those adjustments change the output of the images you acquire.

Images are never a perfect reflection of reality—they have never been. That’s one of the reasons we have complementary modalities. Why do an MR and a CT, or a CT and an MR? Because they give you two different dimensions of the patient’s reality. Each dimension itself is not enough. Depending on what dose you use, what setting you use, what reconstruction you use, field of view, pitch—all of those factors affect the images you get. On top of that, patients come in different makes and models. We are all different. My needs are different from your needs; your context is different from my context. So in this sea of degrees of freedom, how do I appropriate the right settings for that patient? I think that’s a very complicated question to answer.

I think the answer requires the complementary expertise of a radiologist, a physicist, and a technologist. If any of us think we’ve got it all figured out, I think we haven’t. In order to get this to work, we need to have humility in the practice of our expertise to admit that we don’t have it all. We need each other, and we need to respect one another. Why do we need to work together anyway? Because at the end of the day, there is a patient who is supposed to benefit from the elevated expertise that we claim.

CHRIS ST. JOHN 00:12:00 You’ve used the phrase: “Imaging is about the patient, not the machine.” What does that actually mean for what a radiologist is seeing on the screen, what a technologist is doing at the console, and what physicists are doing behind the scenes monitoring or adjusting protocols? Can you expand on that?

DR. EHSAN SAMEI 00:12:24 It goes back to the phrase I said a little bit earlier: images are never a perfect rendition of the reality of the patient. It is a depiction of it. It corresponds to that reality and is highly relevant to the care of the patient, but it’s never perfect. It’s malleable, changeable, and fluid. That’s another way of saying there are different degrees of freedom. An image is a fluid construct that we capture from the patient based on how the imaging is set up, and we want to do right by the patient.

The practice of medicine, though, is often not done in a patient-specific manner; it’s done in an aggregate manner. Let me give you an example. When you go to a physician and they want to do some procedure, they might say there is a 5% chance of this, a 14% chance of that, and a 20% chance of this, so they suggest course B as opposed to course A or C. These are typical things you hear from a physician. What does 14% mean? Does that mean only 14% of me gets affected by it? No. That means there’s a hypothetical population where 14% of that population would be afflicted by the side effect I’m talking about. So essentially, I’m not seeing you—I’m seeing you as an aggregate, as a representative of a population.

There is nothing wrong with that because that’s how we practice medicine. We always average. We look at mean values and confidence intervals. So our care comes from aggregate data, but that aggregate data needs to be applied to the individual patient. That individualization process is something that can be significantly helped by understanding the nuances of image acquisition and understanding the nuances of the patient. That’s where personalized care meets reality.

Ultimately, we want to do right by the patient in front of us, not an average patient. There is no such thing as an average patient—I have not met one.

CHRIS ST. JOHN 00:14:55 How do you even begin to think about approaching that? I think it is very safe to say that in a lot of locations across the country, a lot of folks are referring to manufacturer-suggested protocols. They’re following the guidebook they were given. But because of that, it’s not like competing vendors are constantly talking to each other, comparing their doses, comparing their image quality, and sharing all of that with each other. Because of that, we get a lot of variability across different technologies. I’m curious: how much of what happens at the console is really being driven by the vendor versus the physicist, and how do we reframe it towards the patient?

DR. EHSAN SAMEI 00:15:50 I think manufacturers, rightly so, suggest protocols to be used for imaging patients. You would imagine that’s the right thing to do because, ultimately, they make the machine and understand it best. But those choices are often informed primarily by the attributes and specifications of the machine itself, and also by aggregate feedback that manufacturers have received from practices out there. They do their best, they’re doing an ethical job, and there is nothing wrong with that process at all.

CHRIS ST. JOHN 00:16:32 Oh yeah, I’m not trying to poo-poo manufacturers—sorry, guys!

DR. EHSAN SAMEI 00:16:37 It’s actually great. But ultimately, manufacturers, by virtue of being manufacturers, are oriented toward the products themselves, while medicine is about the patient. Another way to think about it is that the job of a physicist, or a clinical institution in general, is to make sure the diversity of products available at that facility are somewhat commodified—commodified is a negative term, but perhaps makes sense here—so it reflects what the patient needs for their individual care.

I should not care as much whether I’m using a Siemens scanner, a Canon scanner, a GE scanner, or a Philips scanner. Regardless of what scanner I use, the scanner needs to do what the patient needs. In fact, manufacturers will also say that. If you show me a picture of your amazing vacation in Hawaii, my first question shouldn’t be, “What camera did you use?” That camera needs to be transparent. Ultimately, it needs to be about the patient.

Because there are too many degrees of freedom, as I mentioned, those degrees of freedom complicate the process because different devices have different degrees of freedom. When I started at Duke, there were two manufacturers’ devices, and radiologists were in Camp A or Camp B. They said, “We hate the images from the scanners in Camp B,” and some people hated the images from Camp A. I didn’t understand why. I asked, “Why don’t you adjust the settings of the two machines so they are similar to one another?” In fact, that became one of the first projects we did at Duke: adjusting the manufacturer settings within the available constraints so images provided what they were supposed to. If I need to see a liver lesion, measure the size of a lung tumor, or measure stenosis in cardiac disease precisely, how can I adjust the settings available on two different machines—which are named differently, by the way—so I can deliver that outcome? That way, the radiologist looks at the patient, not images from Manufacturer A or B. That is the essence of patient-specificity, patient-centric care, and the specific role of a physicist.

We should start with what manufacturers recommend and adjust so they can meet that ultimate goal. Why are protocols so variable across the field? First is different makes and models of machines. Second, different manufacturers recommend different things because they prioritize different things. One manufacturer might prioritize temporal response, while another might prefer spectral response, or vice versa. But there is also a degree of agency among the users. We like to fuss around with the options available to us; otherwise, we don’t own it. Nobody likes to be told, “This is the machine, and that’s the way you’re supposed to use it.” Most thoughtful people—physicists, radiologists, or technologists—want to fine-tune things because they think they can make it better. That’s a good thing, because sometimes amazing things come about that manufacturers, radiologists, or physicists hadn’t foreseen. That level of customization or accommodation per personal preference is within the landscape. That’s why protocols at Mayo Clinic are different from protocols at Duke or Wisconsin, even for the same machine.

On top of that, sometimes you might want to deviate from the guidance you developed because a particular patient has a specific condition. For example, a liver transplant patient comes back with repeated infections. That is the context of that patient’s care. Should I use the typical liver protocol the manufacturer recommends, or make accommodations for that specific patient? That accommodation is appropriate as long as I know what I’m doing. That adds some degree of variability in how we practice medicine, but that variability is appropriate because it’s done for the personalization of care for that individual patient. We are doing right by the patient.

In our attempt to quantify medicine and reduce variability, I would argue that some variability is bad—for example, using different protocols for patients with similar attributes—and we need to minimize that. But certain variability is good because we are accommodating what the patient needs. Some degree of variability in protocol setup is called for. The job of scientists—physicists and radiologists—is to sort out good variability from bad variability, the right protocol from the wrong protocol, so we can appropriate resources for the best care of the individual patient in front of us, not the average patient.

CHRIS ST. JOHN 00:23:25 I want to touch on this side of the conversation, which has a layer of idealism baked into it to some degree. Not being dismissive, but I’ve listened to you speak tons over the years, and I’m thinking back to a slide you have about all the different qualities that apply to different patients. It was a big number—like 50-odd signifiers or something. Do you remember what that number was off the top of your head?

DR. EHSAN SAMEI 00:23:26 I don’t remember.

CHRIS ST. JOHN 00:23:27 Right, but it’s a lot—race, size, height, different aspects of medical history, a seemingly unending list of all of these things that make us different. If you answer every question on the list, you don’t have an average patient; you have this patient, and you have all the information you would need to image them ideally.

So in this ideal structure, I’m curious what the organizational workflow would look like. And then after we touch on that, I want to pivot to the other side of it, talking about capacity and current-day volumes.

DR. EHSAN SAMEI 00:24:25 The slide you were talking about is a depiction recognizing that the context of care for different patients is different. Whether a screening exam is a pediatric exam, a follow-up exam, a chronic exam, or a transplant exam, each one is a different context. Whether the patient was a previous smoker or a non-smoker matters. If a patient is sensitized and wants to make sure any additional growth is not missed, while another patient says, “I’m afraid of radiation; I want to make sure you use the minimum radiation possible”—all of these are related to the context of care. By the way, the patient’s voice is also significantly important in this space. I don’t want to ignore that. After all, if I’m imaging you, Chris, this is your body. I have to listen to you. You can’t say, “We’re going to ignore you because this is the right thing to do.” You can’t do that.

CHRIS ST. JOHN 00:25:19 Once again, I’m also the worst imaging patient, because I want that dose up! Jokes aside, though.

DR. EHSAN SAMEI 00:25:39 I’m actually in the same zone, by the way! But anyway, see how a physicist approaches this. A non-physicist will say, “Well, patient details matter, and you have to adjust things according to the patient’s needs.” For that, you don’t need a physicist—you can just do adjustments as you go. But you also don’t want anarchy in the space; there is a way to organize it.

So we said, “How about we identify 150 to 200 labels or tags that we can associate with each case?” Now I am taking the attributes of the patient as I understand them from the medical record and associating the patient with a series of tags. If I know the set of tags, I can aggregate different patients together in a smarter way. For example, I can put all critical care patients on a kidney transplant list into a subcategory of its own and define what is most appropriate for that subgroup, as opposed to bundling all of them together and calling it an “abdominal CT.”

That is a way of making personalization, accommodation, and contextualization systematic. This is a good example of how a physicist meets a medical need: something that was subjective or arbitrary is quantified, categorized, and streamlined through a systematic process.

CHRIS ST. JOHN 00:27:23 Thinking about that in practice, management of all of these tags and the variability in imaging as a result feels like a gargantuan project. Even where things are now in imaging, we still have all of this massive variability. We have diagnostic reference levels—a little plug for the new Radiology publication published by Kunal et al., which Dr. Samei is on as well, coming out with new DRLs and giving facilities the opportunity to see what everyone else is up to. But thinking about the scale of what you’re talking about is massively intimidating. Adding all of these tags to an already rich and complicated imaging process, how do you consolidate that from your perspective?

DR. EHSAN SAMEI 00:28:31 I don’t think it’s super complicated, because the data is already at our disposal. The electronic medical record is a rich space where data is present, and AI sorting resources into different buckets is a process that can easily be done. I don’t think it’s super complicated.

I’m still advocating for DRLs based on what I call macroaggregates, but what I’m advocating for here is microaggregates based on smaller tags—a smaller aggregation based on tags. We need to start with the more generalized descriptions we have right now and gradually, over time, stratify the data in a more refined fashion. That’s all it is.

Meanwhile, because we are not there yet and need to accommodate patient cases as they come, we need to think of what we do in terms of guidance, not law. You can think of the practice of medicine on a spectrum: a legal approach saying, “This you shall do; this you shall not do”; guidance saying, “We recommend you do it this way, but use your own judgment for what is best for the patient”; or a super-personalized fashion where a specific thing needs to be done for that patient. I think the third one is difficult to achieve at the moment because the system we just described is not there yet. But we should move toward the middle ground where medicine becomes more guidance-based rather than legal-based.

Personally, I do not like the idea of thresholds—the idea that anything above a threshold is bad and anything below a threshold is good. Thresholds tend to be monodimensional—like, radiation dose above this is bad, dose below this is good. It gives a false sense of confidence that if you are below the limit, you have done right by the patient, which might not be true because it’s monodimensional. It’s like saying all shirts that are less than $40 are good and all shirts more expensive than $40 are bad. You can’t do that—it depends on what you’re getting.

That’s an example of a monodimensional approach. Care needs to be a multidimensional approach and guidance-based. What I’m saying isn’t very revolutionary; we are already practicing human judgment everywhere. We accommodate people and adjust. We need to allow that adjustment. That doesn’t mean throwing away legal recommendations, but we need to give people agency, respect that they are experts who know what they’re doing and are trying to do right by the patient. We must provide professional guidance, but also provide flexibility along with it.

Questions from This Episode

What is the clinical role of a medical physicist in radiology?
A medical physicist brings quantitative and analytical expertise to image acquisition and formation. Working with radiologists and technologists, the physicist helps manage scanner settings, image quality, dose, and other technical variables so the exam supports the clinical question for the individual patient.
Why should imaging teams adjust manufacturer-recommended protocols?
Manufacturer protocols are useful starting points based on equipment specifications and broad user experience. Different scanners and patients create additional variables, so clinical teams may need to adjust those settings to produce consistent diagnostic results and respond to a specific patient’s condition.
Why are fixed dose thresholds not enough to judge imaging quality?
A single threshold considers dose without showing whether the image answered the clinical question. Dr. Samei argues for guidance that weighs dose together with image quality, diagnostic utility, patient attributes, and clinical context, while still giving qualified professionals room to use judgment.