Rethink Imaging

EP 17 • April 17, 2025

Could Our Obsession with Dose Be Risking Lives?

Featured Guest

Dr. Ehsan Samei, PhD

Chief Imaging Physicist, Duke University Health System; Reed and Martha Rice Distinguished Professor of Radiology • Duke University Health System

Listen Now37 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

Radiation dose is the easiest thing to measure in medical imaging, and that is exactly the problem. In part two of his conversation with Chris St. John, Dr. Ehsan Samei argues that judging a radiology department on dose alone is like ranking every restaurant in New York City by the calorie count of a standard dish: the number is real, but it tells you almost nothing about quality. Samei walks through the measurands that actually describe what an image delivers, resolution, noise, contrast, motion, and artifacts, and explains why each one has to be tied to the clinical task the scan was ordered for.

The second half gets specific about what happens when the wrong metric wins. Samei points to fluoroscopy programs graded on exam length, a rule that punishes the thorough care a very sick patient needs, and to CTDI, a dose index that reflects scanner output rather than the radiation burden on a particular patient’s body, age, or sex. He closes with the strongest evidence for imaging’s value: screening mammography finds about three cancers per thousand women screened, and it has driven a dramatic drop in US breast cancer mortality, a benefit he says outweighs the radiation risk by orders of magnitude.

Host

Chris St. John

Host, Rethink Imaging • Imalogix

Featured Guest

Dr. Ehsan Samei, PhD

Chief Imaging Physicist, Duke University Health System; Reed and Martha Rice Distinguished Professor of Radiology • Duke University Health System

Watch the Episode

Key Takeaways

Dose is the calorie count of imaging. Samei says ranking radiology departments on dose alone is like rating New York restaurants by the calories in a standard dish, and calls the current dose-only scoring of departments “a joke.” Calories matter, but no one would hand out Michelin stars with them.
CTDI, the standard CT dose index, measures scanner output, not patient burden. Body size, which organs sit in the beam, age, and sex all change actual risk, and today’s metrics ignore every one of them. Dose needs the same personalization as image quality.
Bad metrics actively harm patients. One entity graded fluoroscopy quality by exam length, but a genuinely sick patient may need an hour under the beam, so the metric penalized proper care. Samei’s rule: quality measures should be guidance, not thresholds.
A trustworthy image quality measurement has to clear three bars: peer-reviewed validation, ownership by the whole community (the AAPM stakeholder roundtable, published in the Journal of the American College of Radiology, brought physicists, radiologists, government agencies, and patient advocates to the same table), and demonstrated correlation with what radiologists actually detect. ICRU Report 54 established task-based measurement as the standard.
The benefit-risk framing is lopsided. Screening mammography finds about 3 cancers per 1,000 women screened and has driven a dramatic drop in US breast cancer mortality, while countries running mammography without tight quality management have not seen the same benefit. Samei compares the radiation risk to pennies set against thousands of dollars of benefit.

Full Transcript

Ehsan Samei Part 2 Final Transcript

[00:00:02] Intro: Welcome to Frame by Frame Rethink Imaging, a podcast by Imologix. Here, we explore the intricate world of medical imaging, aiming to dissect the field and inspire both professionals and curious minds alike. I’m your host, Chris St. John. Welcome back to Frame by Frame Rethink Imaging. Today, we are picking up part two of our conversation with doctor Hassan Sameh. If you haven’t listened to last week’s episode, part one on why medical imaging matters, we recommend going back and taking a listen before joining us today.

[00:00:37] Chris St John: In today’s episode, we are going to explore how we approach measuring the benefits of imaging and why those measurements are important to talk about. And we were just on a little break from our conversation, doctor Samei, and getting not ahead of ourselves, but just, you know, we were chatting away. And I wanna jump right back into what we were talking about on the break, which is you mentioned measurands versus measurements. And can you just explain to me and our listeners the difference between those two concepts to get things started?

[00:01:04] Dr. Ehsan Samei: Thank you again for doing this. You’re such a great conversationalist, and I appreciate the the time that we have spent together. So measurand is an attribute worth measuring, and measuring is what we do. I’m a physicist. I’m a medical physicist. We like measuring stuff. So we measure measurements. So, for example, you can think about what are the attributes of air, humidity, temperature, pressure. These are measurements, and then we can apply our measurements to measure those things.

[00:01:34] Chris St John: Okay. Super simple and straightforward. I got that far. Then what are the measurands when it comes to not just imaging, but the benefits of imaging? This is where it starts to get super murky again.

[00:01:49] Dr. Ehsan Samei: You know, when you think about the way we practice science, science is primarily is a quantitative enterprise. A lot of the work that we do in science is quantitative. You know, in the enlightenment, we divided the reality into qualitative and quantitative. And quantitative took off. And with quantitative, we created industry. We created the science. Modernity and the approach to the universe in terms of measuring the universe, assessing the universe, improving the universe, a lot of them are numerically based. So we’d like to measure stuff. Why do we measure stuff? Because if you measure them, then you have a sense of control over that attribute. You can target certain things. You can track them. Like, for example, you can how do you know you’re overweight, you’re underweight? You measure yourself. You see, oh, the weight has gone up gives you something to gauge yourself as opposed to how I feel today. I’m not sure which way, by the way, is better. But the point is that having a measurement at your disposal to figure out how you’re managing your health is helpful. Would that scale give you everything that you need to know about your health? No. Or all the measurements that we make on your blood in terms of how much cholesterol you have and lipid, all that stuff, do they fully describe your blood chemistry? No. But it’s very helpful. We can practice medicine with it. So when it comes to imaging, it’s the same story. In order to be able to practice and target radiology to a certain level of quality, I need to have a measurands at my disposal to be able to target them. Do I fully characterize the benefit of imaging? No. But having some numbers would be very helpful as long as numbers are not misleading us. Right? Sometimes you can have a fake number or number that is not really representing the reality. So any sort of measurements or measure that reflect the measurand is we try to represent numerically an attribute of that entity that is helpful and reflective to understand that entity. So Right. Going back to your original question, do I can I measure the benefit directly? No. In fact, really, you can’t measure anything directly, but I can indirectly measure it with a series of measurements. Right.

[00:04:16] Chris St John: Because in medicine, we’re tracking outcomes by, like, the success of treatments, but imaging in itself is diagnostic. So how do we connect measurable outcomes, like lives saved, complications avoided, and connect it to the diagnostic tool?

[00:04:36] Dr. Ehsan Samei: Yeah. Right. So you’re asking the right question. First of all, in medicine, our goal is to provide health or advance health, to foster health. So, ultimately, quality of life, you know, years of life, and mortality are the proper measurements that we could should go after. Often, those things cannot be measured now right away. Why? Because first of all, it takes a while to get those results. Now, by the time you pass from my intervention or not pass from my intervention, I’m long dead myself. So it doesn’t so often we cannot get our hands on that thing ourselves directly. Secondly, sometimes things get very compounded. Because I don’t know, you might not because of what I did to you today, but somebody else did something to you six months from now. And then now I cannot really that number that I was planning to get out of you as a patient is now contaminated. I can’t get my hands on it. I cannot put you in a test tube. You are gonna live your life, and you’re gonna be affected by a variety of factors, which is different from you to me, to my wife, to my daughter. They’re all over the place. So this confounding effect makes it very difficult to try to actually measure any of those things directly. So when it comes to diagnostic care, that would take a few step back and say, okay, I can’t measure that very easily. Is there something that I can measure? I would say there is one thing you can measure. You can measure radiation dose, which in fact, that’s probably the easiest thing you can measure, which the problem with radiation dose is that as we talked in our previous podcast, is misleading because you only reflect one attribute, one measurement of that process. And in fact, it’s not even a directed to benefit at all. Right? But the other thing that images are supposed to provide, when you think about why do we do imaging, we do imaging in order to provide information from the interior of the patient. So I can come up with some measurements that reflect how well the information is reflected in the images that I’m given. So if there was something there, how truthful the image represents that reality. So information transference factor, if you will, right? Did I get that? If there was something in there, did I get it in my, did it get captured in my image? Right? That is an indirect measure of the benefit that the images themselves supposed to provide. Now, how that benefit get propagated to the life of the patient and the treatments and other confounding effects, I cannot tell about that. But I can tell you whether images were any good in representing the content that they’re supposed to produce. Am I making sense?

[00:07:15] Chris St John: Yeah. Yeah. Absolutely. You’re talking about dose, which is the most easily measured measurant, but also saying that it’s an unreliable measurant. Is that correct?

[00:07:29] Dr. Ehsan Samei: Unreliable or not misleading might is too monodimensional. Monodimensional means, like, imagine for a second you’re trying to assess whether food is good for you, and the only measure at your disposal is the calorie. That’s the only thing you measure. Right? You’re actually a food expert yourself.

[00:07:46] Chris St John: Can you expand on that a little bit?

[00:07:48] Dr. Ehsan Samei: I mean, if I tell you that, essentially, you can rank order restaurants across New York City based on the calories that they put out there for a given standard dish, do you think that would be a good reflection of the quality of the food that you receive at those restaurants?

[00:08:04] Chris St John: Absolutely not.

[00:08:06] Dr. Ehsan Samei: Yeah. So why would that be? Do would you say that calorie is irrelevant? You probably would not say calorie is irrelevant. It’s a measurement. But if that’s the only thing you got, there is no other measurements at your disposal. And you measure the quality of my hospital versus another hospital, my restaurant versus another restaurant based on only that one single metric. I think you can’t really mislead yourself. No. Imagine Michelin stores would be based on the calorie of the restaurant. That would be a joke. Right? Sometimes the dose is so easy now. I feel like now we have set up a system, at least in this country or many other countries as well, that we just measure that calorie input or output, whatever you want to call it, of imaging as a metric of how good you are as a radiology department. That’s a joke if you ask me. That doesn’t mean you should ignore the calorie, but you should think about, okay, can we say more? Is there other measurement? Can I measure the sourness or sweetness or bitterness or whatever? Are those things relevant?

[00:09:10] Chris St John: Right. So, I mean, in terms of imaging then, like, if we’re talking about noise and image quality and and these other things, like, at what point does do we start to have a clearer picture?

[00:09:23] Dr. Ehsan Samei: This is a science of image quality. Right? I think we what we want to do in image quality

[00:09:29] Chris St John: I was saying clearer picture, uh, metaphorically, but then realized it was applying quite literally and laughing to myself about it.

[00:09:36] Dr. Ehsan Samei: It is literally. That one is well intended. Here, I wanna know if this picture is good and that picture is bad. The picture that is good is a picture that represents the reality more faithfully than the picture that is banned. Okay. So when you say faithful representation of reality, faithful representation of information, translation of information from a real domain to an image domain. Right? What are the things that you consider good quality? Like, there are certain things that you know, for example, you wanna have high resolution. We already know that. Like, for example, if you get a latest iPhone, it claims to have the more pixels representing that image. And more pixels, that means the pixels are actually tinier and smaller. That means the images are higher resolution. That means the spatial details are more faithfully reflected in the image. The other thing is the noise. It means apart from those spatial details, what are the fluctuations that are existing that manifest themselves primarily, by the way, when you in a photographic imaging, when you take an image without a flash. The sharpness, the clarity is not there because of the statistical fluctuations of pixel. So resolution is one. Noise is one. Contrast. So if I have a black region and a white region, do they represent themselves differently? That would be yet another one. Images in imaging, patients perpetually move. We have we are dynamic entities. Our heart is beating. We breathe. There are internal movements within the body itself that we do not control. And when we take a snapshot when you take a single snapshot of the patient, that motion is get solidified into the images. So all images have certain amount of motion artifacts in them. Some of them are large. Some of them are small that we ignore. So motion would be another one. There are when it comes to, like, CT images, there are all sorts of other artifacts like streaking artifacts that are present. There are blooming artifacts that are present that would mask reality of what is going on. So these are attributes. Do they reflect the actual quality of the images perfectly? The answer is no. No measurement reflect everything perfectly. That’s my point. But, approximately, yes. In the same way, a food package that has calories but also saturated fat and salt and sugar and all that stuff, that label is better than the label that only has calorie. So I think what we need to do is to identify attributes, measurements that reflect that clarity that you mentioned earlier, and an image that has produces higher clarity, then that image would be a better image. Has a potential to provide higher benefit to the patient and to the practice.

[00:12:30] Chris St John: Right. So are there ways to measure a measurant of quality in radiology? And how do we know if the measurants are relevant or reliable?

[00:12:45] Dr. Ehsan Samei: Right. So that’s a great question. I think because the science of image quality is, uh, is still developing science, I think whatever measurement that we make, Chris, needs to lend itself to scientific scrutiny. By scientific scrutiny is that in science, if I come up with an idea, you don’t take my words for it. I have to prove to you that what I have is right, and I have to prove to my peers that what I have is right. So peer review is a hallmark has been a hallmark of scientific inquiry. So a measurement needs to stand the scrutiny of a peer review, needs to be proven by others other than the person who developed it, in my opinion. And community at large needs to accept it. So it’s not just for me or for you or for this company or for that university to say this is the right way of measuring noise or resolution or whatever. I think we need to communally come together and say this is the right way of measuring it. The third thing I would say is that we need to have some validation that measurement that we have come up with or the way that we have done the measurement is relate to that outcome that you refer to early on in this conversation. And remember, the the measurement itself, if I measure noise and resolution, is not what images are all about. Radiologist don’t go and measure noise in images or measure resolution. The radiologist would detect the cancer if the cancer is present. So with those measurements of mine that I just made, are they related to the likelihood of the practitioner being able to do the diagnosis that is needed? If my measurement is not related, cannot be correlated with what the physician is able to do with the images, my measurement is useless. Right? So it needs to be validated to a clinical touch point. Measurements are physical stuff. They’re geeky stuff by themselves are useless. It needs to be anchored to clinical reality. So it needs to be anchored to clinical reality. It needs to be scrutinized, needs to withstand the scrutiny of the peer review, and needs to be owned by the entire scientific community. That’s the only way I can trust a measurement.

[00:15:08] Chris St John: So how do we do that?

[00:15:09] Dr. Ehsan Samei: Well yeah. That’s again, we right. How do we do that? First of all, I think anybody who comes with the measurements, I think we should see if there is evidence that there is some confidence for that. There is a peer review publication that goes with it, number one. And secondly is that the variety of stakeholders can accept it. Right. So AAPM, I think, a couple of years ago, put a panel together, and the result of that panel, uh, was published. It was not a panel. Sorry. I’m misspeaking here. There was a stakeholders roundtable in which AAPM is stands for American Association of Physician Medicine. It was not the AAPM initiative. It’s AAPM just invited a whole bunch of other entities, including radiological societies, nonradiological societies, governmental agencies, even patient advocates around the table and say, how do we ought to measure quality of imaging? How do we measure? And we realized that we need to share that space. I can’t physicists cannot dictate this. Radiologists cannot dictate this. Government cannot dictate this. It is owned by all of us together. So I think we need to have an entity that a table, a roundtable, which everybody has the same real estate at the table, that can together come up with a series of measurements on ways to do those measurements that is accepted by all. That way, we can move together. That’s the way AAPM has advocated. I totally support that idea that the result of that roundtable was published in an article in early last year in the Journal of American College of Radiology. I highly recommend your listeners to check that out. I think that’s the right strategy. And the problem is this. Without that, we become up with bad measurements. Early on, one of the measurements and it might still be in practice. One of the measurements that was instituted by by some entities was in fluoroscopy imaging. You know? The fluoroscopy imaging is a form of imaging in which you, like, get a video of the patient. It’s not a single image. It’s a series of images. Right? And usually, when the procedure becomes complicated, the length of the exposure gets longer and longer. You might image the patient for five minutes, ten minutes, twenty minutes, thirty minutes. So one of the entities suggested that the length is a should be a measure of the quality of that practice. Shorter is better. All things being equal agree. A shorter fluoroscopy examination is better than the longer one. But if you’re really, really sick and I have to do an hour of examination, Is that really I’m doing you justice by limiting that to five minutes? No. Right? So the point is that going back to one of the things that I mentioned in our previous podcast, caring for the individual patient is number one priority in any health care enterprise. And when you throw in a bad metric for quality and measure hospitals based on the length of the fluoroscopic examinations, undermines the actual care, good quality care for the individual patient. So bad measurements is bad for practice. So that’s why it’s important to make sure the measurement that we come up with are owned by the entire community.

[00:18:34] Chris St John: And just to build on that, I mean, it ties back to what you were talking about, a threshold versus a guidance. Right? Like, this is what you’re building on.

[00:18:42] Dr. Ehsan Samei: Yeah. Owned by the community and should not be threshold based. Should be guidance based. Guidance. So that means you are honoring the reality of the need for personalized care. We’re not saying you need to practice medicine haphazardly, but you need to allow the agency and expertise of the practitioner and the voice of the patient to be present in the care process.

[00:19:10] Chris St John: So we’ve talked quite a bit about personalizing the imaging process, but how does personalization apply specifically to radiation dose? Should dose also be tailored individually for each patient?

[00:19:24] Dr. Ehsan Samei: Yes. That’s a good question, Chris. As I have talked in the first video, first podcast and also in this one, big part of what we need to focus on is a personalization itself, whether in the context of image quality or radiation dose. We need to customize our radiation dose and image quality to the individual need of the patient. So if you are dealing with something that is related to your thyroid or your liver or your brain or your heart, We need to make sure the imaging that we do specifically extract the information needed for that specific care that you’re coming to the medical facility for. We also, at the same time, need to minimize the radiation dose that you receive. You know, radiation is a gift to medicine, but just like any other gifts, uh, including pharmaceuticals and so on, you need to use it responsibly. What is the right dose? How do we measure the dose for an individual? By and large, the medical community has taken this aggregate approach toward radiation dose that doesn’t personalize it. So for example, in the context of CT, we use CTDI, CT dose index as a metric of radiation dose to the patient. But actually, that is not the metric of radiation dose or burden of radiation to the patient. It’s rather is a reflection of the output of the imaging device in terms of radiation that puts into the patient. So the level of harm, hypothetical, theoretical, lean, or threshold oriented, however it is calculated, needs to be related to the individual mister x and missus y. What part of my body is exposed matters. How much radiation flux goes to different organs of my body actually matters. What is the size of my body that matters. What is my age matters. If I’m younger, if I’m older, changes the calculation of risk completely differently. What is my sex affects my radiation burden. All of those factors are by and large are ignored in the measurements that we are currently using for radiation dose in medicine. So radiation dose in medicine also needs to be personalized in the same way that image quality is personalized.

[00:21:56] Chris St John: Okay. And let’s pivot a little bit. Right? We’re talking about measurements. We’re talking about measurements and the challenges in measuring the quality of images or imaging. But how do we tie the results of imaging in this clinical chain of practice to the result. Right? So for instance, if you have a trauma patient and you identify, like, a brain bleed, or if you identify cancer in a patient through imaging or whatever it is, you are branching out through this health care system into different departments, into different areas. And when you’re trying to measure all of that, I feel like it starts to get incredibly convoluted incredibly quickly.

[00:22:47] Dr. Ehsan Samei: That is true. And that’s one of the reasons that, you know, at least in radiology, we try to measure at least that’s an easier task of answering the question, do images provide the information that they’re supposed to provide? So there are a variety of doing that. Uh, one is that you measure things such as we talked about this resolution and noise and contrast and motion and blooming and streaking and so on. You can measure those things. Those are equivalent using the food analogy, equivalent to sourness and sweetness and saturated fat and so that we measure those things. And those are what in imaging language, we call them task generic. They’re attributes of the images in representative ground truth that might be present in the patient. Then you can translate those numbers into what we call task specific reality. That means, now if you happen to have a bleed in your brain or plaque in your heart or cancer in your liver, how confident we are that image can represent that abnormality well. That would be task specific representation of those attributes. And by the way, that task is specific. This has been that was established by a international body many years ago. ICRU report number 54 says, if you wanna measure image quality, that’s the right way of measuring image quality, by the way. So there have been a fair amount of science to translate these noise and resolution and motion and stuff into that task based performance. If I know the resolution and noise and those attributes, I can put them into a formula, if you will, and tell you how well this image would represent brain bleed versus cancer. Then depending on the indication for which the patient get imaged, which is often we know that. Right? Nobody go and images you because you wanna get an image. Usually, that’s bad practice. Right? There is always an indication. There is a reason we’re suspecting of lung cancer, we’re suspecting of breast cancer, we’re suspecting of this or that. So there is a target in mind. If you give me the target, I can tell you that image is good for that target. Or the dose need to be adjusted up and down to make that target more visible to the level that the level of visibility is sufficient for the physician to act for your care. Right? You know, we don’t need to know everything. We need you to know enough to what we need to do with you clinically taking the next step. That is the unique guidance that is provided. So we have done, at least in my laboratory, we have done a number of studies. And I at least I have done three studies in my lab. These are studies have not been very many of them, That we have done this. We have done the translation from those basic measurements to task specific performance, like detection of cancer. And then we have done the detection performance with actual practicing radiologists and ask the question, do my prediction of that task performance measures against the practice itself? When I say validation, that’s what I’m talking about. It means now my measurement is now made relevant. I’ve demonstrated that what I’m measuring is actually relevant. Would be equivalent. This is almost ridiculous what I’m saying, but you get the idea. Imagine if I come up with a formula that get the label on the back of the food packaging, add it to my formula, and tell you how delicious the food is. Imagine for a second. This would be task based performance type of thing. And then do a study in which I ask a whole bunch of people to eat a whole bunch of food and measure their deliciousness, rank all the deliciousness, and see how my formula matches theirs. If they match, then you’d believe my formula. Right? That’s basically like this. I would say, though, when it comes to image quality, I think our formulas are much better than using the food analogy. Food is really much more complicated.

[00:26:48] Chris St John: Right.

[00:26:49] Dr. Ehsan Samei: Because, you know, medical images at the end are scientific technological tools. So we are a little bit more lend themselves more to that kind of measurement while, like, food, which I’m using that analogy over and over, is more of a artistic reality. Right? Yeah. I cannot come up with an algorithm to tell you this take a picture of your painting and say, okay. This worth hundred million dollars based on my formula. That doesn’t work that way.

[00:27:13] Chris St John: Right. That would just be the golden ratio at that point, which it would be a great little cheat code if you could just paint a painting with that ratio, and then it would just sell for a bunch of money. That’d be great.

[00:27:22] Dr. Ehsan Samei: No. No. Never. Right. I mean, yeah, you could ask the question.

[00:27:27] Chris St John: Is the question should radiology departments be doing more to document outcomes and demonstrate their role?

[00:27:34] Dr. Ehsan Samei: Yeah. I think we should do more this kind of measurements than describing. I think that needs to happen. I think we need to also measure the outcome. All of measurements that we make are insufficient, but we need to have measurements nonetheless. In fact, every form of measurement is a form of approximation. Right? And I think given the fact that nowadays people are just so consumed by those or obsessed by those, I’m just about to write. I think we need to go out of our way to demonstrate the value even though it’s so obvious to us how well the images represent what they’re supposed to represent and what would happen to the patient if those images are not acquired or suboptimally acquired. I think we are very fear minded in the way we practice medicine. Again, we need to make sure we maintain utmost safety in the care of the patients. There is no question, but we should never lose sight of the fact that there is a significant amount of value and benefit to take care of individuals that imaging provides, and you never want to compromise that.

[00:28:41] Chris St John: Okay. So let’s play around a little bit, if that’s okay. Let’s say theoretically, you had unlimited data and unlimited funding. How would you design a study to answer the question, how many lives does imaging save? Just real quick on the spot, just solve this problem for us.

[00:29:05] Dr. Ehsan Samei: I think what I would do first of all, number one, I keep track of the indication for which the images of a certain series of patients were acquired. So let’s say we would get, you know, randomly selected thousand patients that were imaged at my facility and asked the question, what indication brought them to that scenario? That will keep track of actually what the images provided for those individuals for that scenario. What I would do is that I would ask the question from the practicing physician who care for the individual. If that image was not available, how the course of the treatment or follow-up for the patient would have been different? And you make a predictive modeling, what would happen to that patient, Not a hypothetical patient, that particular individual. I also asked a question if the images were, like if radiation dose for that patient was reduced by half or increased by factor two, either one. Now I can do predictive modeling and see how much the quality of those images would have been better or worse, how much uncertainty or error might have been induced by suboptimal imaging, and also follow that in predictive modeling. Now I can know for sure what would have happened if those images for those 1,000 patients were not acquired or suboptimally acquired. Now I can, you know, clearly say for mister x, for missus y, what would have happened with or without imaging, with good quality imaging and bad quality imaging? Would somebody pay for such a project? Again, I feel like it’s such a dumb thing to pay for this project. That’s the problem if you haven’t done that. Because the odd benefit is so obvious. Nobody wants to pay for something you already know. Right? That’s why we haven’t funded it. And as a result, we have gotten this lopsided idea that calorie is the only measure of quality of the food, and dose is the only measure of quality of imaging. That’s the reason.

[00:30:57] Chris St John: And so then how do you feel about the approach of quality measures being so based in dose?

[00:31:04] Dr. Ehsan Samei: I think dose needs to be one of the quality measures, but one of the minor ones that we have to keep track of, but not the only one. In the same way, for food, again, I’m using a food analogy, we need to measure the calorie intake of the food. That would be helpful. But we do need to include other measurements of quality. Again, I will mention resolution. I will mention noise. I will measure motion. All of those measurement needs to be done in the context of the indication for which the image was being acquired. So if you do a cardiac exam, then those measurement needs to be made relevant to cardiac imaging, not some generic number out there. It needs to be relevant to that, what they call task based performance. So the numbers need to lend themselves to the task based prediction. Because at the end of the day, task is what matters. Nobody cares about the resolution noise of the image. We care about whether I can see the abnormality present, and can I adequately quantify it and characterize it so to know what to do with the patient? I think, again, making sure the benefit is not compromised is something that has not been on the forefront of people’s mind. We have perpetually tried to reduce the dose. And maybe some dose should have been reduced because they were too high. But at the same time, if we have that as the only measurant, we might compromise the patient care. That’s what I’m worried about and all of us should be worried about.

[00:32:39] Chris St John: And so we are getting close to the end of time for the second piece. Is there anything you want to touch on or address before we sign off?

[00:32:53] Dr. Ehsan Samei: One thing I just wanna again, on the benefit side of imaging, early on, we talked about, you know, how do we know the imaging provides benefit. Right? I use the analogy of exploratory surgery, an analogy, the practice of exploratory surgery. And I just also wanna highlight a couple of, you know, landmark studies, one that was related to breast cancer detection through mammography, and the other one is lung cancer screening. Both of those are now standard of practice. I, um, recently participated in the a medical conference in Caribbean region, and they were talking about how how the benefits of mammography has not manifested itself in those Caribbean countries. Here in The US, since the screening mammography became the standard of practice in this country, there have been a dramatic reduction of breast cancer mortality in The US. Dramatic. So in fact, people argue that the reduction is primarily due to the mammographic mammography screening practice. So there is no doubt that the imaging that we are doing in breast care has significantly reduced the mortality of breast cancer. And the quality that those images produce, and we, you know, we we measure the quality of their demographic exams like a hawk. The quality is really, really high. It has really made a difference. In the Caribbean countries, they’re also doing mammography. But because the quality is not as closely managed, they have not seen the benefit. So this I provided an example because demonstrate not only that imaging matters, but also quality of those images matters as well. Now you could ask me the question if you’re fearful of the radiation dose. You can say, wait a second, doctor Sameh. We are exposing all these patients. What about that dose? Do are we causing breast cancer? Can you tell me about that? So I will tell you the mortality has decreased in The US. But, you know, from one thousand women who get screening mammograms, do you know how many cancer we find in them? Nope. Three. Three out of one thousand. So you could ask me the question, how about the other nine hundred and ninety seven people that got exposed? In fact, they get exposed every year. Right? What about the radiation dose? So this is a demonstration that even though there is radiation dose, uh, for a vast majority of these patients, they’re not we should not call them patients because they’re not patients yet. They’re they’re people. Majority of these women, there is no direct benefit. There is a direct evidence that the benefits significantly outweighs the risk associated with that hypothetical cancer induction that we’re talking about. I can give you more. I mean, the lung cancer screening trial is exactly the same scenario. There is no question that the benefit of imaging significantly outweigh the radiation risk. When we take talk about benefit versus risk, we are misrepresenting. We are assuming there are apples two apples that we’re comparing against each other. They are not comparable. One is in the order of pennies. The other one is the order of thousands of dollars. We are comparing it against each other. It’s not fair. It’s not fair. Yes. We need to reduce the pennies. We need to manage the pennies, but that’s not the only attribute. People do not come to hospital to get dosed. They come to the hospital to get imaged. So if something worth measuring is the quality of those images.

[00:36:24] Chris St John: I feel like that’s a great place to end things.

[00:36:27] Dr. Ehsan Samei: Excellent. Good.

[00:36:29] Chris St John: Doctor Samei, thank you once again so much for joining us here on Frame by Frame Rethink Imaging. It is always a joy to have you here, and I can’t wait until you’re back again.

[00:36:39] Dr. Ehsan Samei: Thank you. Thank you for honoring me with these conversations, and thank you for listening to me. Hopefully, that would be edifying to your listeners as well.

[00:36:47] Chris St John: Absolutely.

[00:36:50] Chris St John: Frame by Frame Rethink Imaging is brought to you by Imalogix. Here, you’ll find engaging interviews with thought leaders, experts, and patients, sharing stories that showcase the transformative power of medical imaging. To discover how is rethinking imaging in health care, visit

Questions from This Episode

What is the difference between a measurand and a measurement?
A measurand is an attribute worth measuring; a measurement is the act of quantifying it. Samei’s example is air: humidity, temperature, and pressure are the measurands, and instruments produce the measurements. In imaging, the measurands that matter are attributes like resolution, noise, contrast, and motion, because together they approximate how faithfully an image represents what is inside the patient.
Why is radiation dose a misleading way to judge imaging quality?
Because it is one-dimensional and not tied to benefit at all. Samei compares it to judging every restaurant in a city by calories alone: the number is real but says nothing about whether the food is any good. He adds that CTDI, the common CT dose metric, reflects machine output rather than the radiation burden to a specific patient’s organs, size, age, or sex.
How do we know an image quality metric actually predicts diagnostic performance?
By validating it against what radiologists can do with the images. Samei’s lab ran three studies that translated basic measurements like noise and resolution into task-specific predictions, such as cancer detection, then tested those predictions against practicing radiologists reading real cases. When the formula’s prediction matches observer performance, the metric has earned its clinical relevance.