Rethink Imaging

EP 53 • September 17, 2026

Updated CT Diagnostic Reference Levels: A 10-Year Analysis of 5.2 Million Exams

Featured Guest

Dr. Thomas Griglock, Ph.D. and Dr. Kalpana Kanal, PhD, DABR

[Dr. Griglock] Executive Director at Lucere • [Dr. Kanal] Managing Director at University of Washington

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

Medical imaging facilities continually balance dose reduction with diagnostic image quality, yet updating national benchmarks historically took years of manual data curation. On this episode of Rethink Imaging, host Chris St. John speaks with Dr. Kalpana Kanal, lead author of the updated national CT diagnostic reference level paper, and Dr. Thomas Griglock, a clinical medical physicist involved in the Imalogix research effort, to discuss an analysis of 5.2 million exams from 2025.

The conversation explores why national DRLs dropped 22% over 10 years while achievable doses fell 9%, pointing to tighter protocol management across the country. Dr. Kanal and Dr. Griglock explain the shift from aggressive dose reduction to clinical dose optimization, including why anatomy, such as the skull, limits dose reduction in head scans compared with body protocols. They also share practical ways imaging leaders can compare internal protocol data against national standards regardless of facility size or software budget.

Host

Chris St. John

Host, Rethink Imaging • Imalogix

Featured Guest

Dr. Thomas Griglock, Ph.D. and Dr. Kalpana Kanal, PhD, DABR

[Dr. Griglock] Executive Director at Lucere • [Dr. Kanal] Managing Director at University of Washington

Watch the Episode

Key Takeaways

National CT diagnostic reference levels dropped by 22%, and achievable doses fell by 9%, over a 10 year period.
A narrowing gap between achievable doses and DRLs indicates tighter protocol management and less variation within imaging facilities.
Dose optimization must remain balanced with diagnostic image quality, especially for anatomy such as the brain where photon thresholds affect gray-white matter differentiation.
Acquisition-level parsing, axial image sizing, and truncation correction enabled automated curation of a much larger clinical dataset.
Facilities can benchmark protocols against national standards by reviewing a small internal sample or using automated dose monitoring software.

Full Transcript

Rethink Imaging Podcast Transcript

Guest: Dr. Kalpana Kanal, Dr. Tom Griglock

Host: Chris St. John

CHRIS ST. JOHN 00:00 Welcome back to Rethink Imaging. Thank you to Dr. Kalpana Kanal for joining us today, as well as Dr. Thomas Griglock. Dr. Kanal is lead author on a recently published paper where the authors redid your diagnostic reference levels and achievable dose paper from 2014. Kalpana, would you mind telling us about that original paper—what a DRL is, what achievable dose is, and why we have now updated it ten years later?

KALPANA KANAL 01:18 Great question. I can talk about the 2014 paper. I was the chair of the American College of Radiology Dose Index Registry. As you know, the registry was very new—it had come out, I think, in 2011. In 2014, after we had enough data, we decided to publish DRLs.

A DRL is basically a 75th percentile of a dose metric. It is a threshold beyond which you want to investigate why your doses are high. It’s not that you fail at the 75th percentile; it’s just a threshold. If my doses are over that, that means I should investigate why I’m one of the facilities that has a higher dose than most of the other facilities.

The 50th percentile, or achievable dose, is where you can realistically achieve. The 75th percentile is a threshold, but ideally your doses should be somewhere between the 50th and the 75th—closer to the 50th. Because of patient size variation, you could be going up all the way to the 75th percentile. As long as you’re in that window, you’re good. But if you go beyond the 75th percentile, you should be investigating why your doses are much higher.

CHRIS ST. JOHN 02:06 Amazing. Can you tell us what we found redoing this paper ten years later?

KALPANA KANAL 02:16 Redoing the paper 10 years later, we do see a trend in a decrease in CT doses, which is fantastic. I believe the numbers were about a 22% reduction in DRLs and about a 9% reduction in achievable doses over the last 10 years.

Why is this significant? Because we do about 90 million CT exams in the United States. Even a small amount of reduction in dose has an impact on your patient population doses because we do so many CTs in the US.

CHRIS ST. JOHN 02:53 If the DRLs fell 22% and achievable dose fell about 9%, what does that gap tell you?

KALPANA KANAL 03:05 The gap tells me that facilities are getting better with protocol management and the variation in the doses is less. So basically, your box-and-whisker plot, your range, is much lower. Even though the doses in the 50th percentile didn’t decrease much—about 9% or 10%—a 22% drop in the DRLs means those ranges are getting much tighter in terms of the protocols.

CHRIS ST. JOHN 03:40 My last high-level question before we start to get down into the weeds: what can different facilities do to engage with this new dataset?

KALPANA KANAL 04:06 Facilities can look at this new dataset and look at their own doses for comparison to see if they are in the ballpark. For example, let’s say a head CT without contrast in the new data is 50 mGy, just as an example. A facility can look at what their doses are using many methodologies. They can have their physicists do a comparison, or they can look at their dose monitoring software and come up with their average and median values for comparison.

If they have none of that, they could even start as small as looking at their last 50 exams for a head CT without contrast, see what their median numbers are, and do a comparison to this paper. That will tell them if they are in the ballpark, if they are over the 75th percentile, and if they need to look at their protocols more closely and start modifying them so they are in sync with the rest of the country.

CHRIS ST. JOHN 05:06 I want to address the size and recentness of the dataset as well. Kalpana, I believe your original study, which was published in 2017 with 2014 data, had about 1.3 million studies. This more recent study has about 5.2 million, all taken from 2025. Tom, would you mind speaking on that a little bit?

THOMAS GRIGLOCK 05:39 This entire project originated over a dinner at RSNA last November, when we had a lot of wonderful physicists and radiologists in the room talking about what we’re doing at Imalogix and our desires to get more into the research space to help academics get research out quicker. The question was posed: what is the first thing that a practicing clinical and academic group of physicists and radiologists would like to see? They said these dose reference levels.

We had already done some preliminary work on this and were able to turn around these numbers. I think RSNA was a little late last year, so around December 2nd, we got some preliminary numbers out and were able to share those analyses from 2025 with this author group. We had a couple weeks before the turn of the new year, and on January 2nd, we said, “Okay, we’re going to rerun all of our data, put it through the same analyses that we have set up, and look at all the data that we have from 2025.”

While the DRLs were published in mid-July of 2026, the DRLs themselves were known to us and the author group as a whole on January 2nd of 2026. We took the next four weeks writing the paper and submitting it. It was submitted at the very end of January, and the rest, as they say, is history.

That got people thinking about what we’re actually able to do with these systems that we’ve built over these last years. We’re talking about real-time benchmarking, and that in and of itself really changes the game. It gives facilities the ability to compare how they’re doing today to how other places around the country are doing today, which is very different.

CHRIS ST. JOHN 07:50 We normally do our best on this show not to mention Imalogix by name, the company that we work at. But in this case, I feel like it has relevance given that we went through Imalogix to get all of this data together. There were some slight differences in approach between the original study and now. Would y’all mind talking about both the acquisition-level approach and the difference in patient sizing in the original paper versus the updated one?

THOMAS GRIGLOCK 08:34 There were differences in the methodology. This is something that we had to work out with the author group at the very beginning to explain how our data varied from the previous data. In the years between the 2017 paper and now, there have been AAPM task groups talking about patient sizing methodologies. We went back and looked at this from the perspective of: what is the state of the art for how you do patient sizing? What is the state of the art for how you decompose multiphase examinations or multi-examination studies?

For example, if a trauma patient comes in and you do a head, a C-spine, an abdomen-pelvis, and a chest CT, previously that big clump of acquisitions would get thrown out because you couldn’t easily tell which doses went with which. The systems that we had in place took care of all that. Our systems would take that head acquisition, split it out, and put it into a head protocol category. It would take that C-spine, split that out separately, and do separate analyses on that.

We can do acquisition-level parsing, so we can include more examinations and didn’t have to throw those out. We also did patient sizing based on axial images, which is a more accurate metric supported by recent AAPM task group reports. That allows us to give more consistent and faithful patient sizing.

Then we also implemented truncation correction. For C-spines, chest PEs, or large patients, anatomy gets cut off because it’s outside the CT field of view. There are published, peer-reviewed methods to correct for patient size even if you can’t see the full size of the patient in the image. We implemented all of those into this paper to make it state-of-the-art.

The result is an updated methodology that aligns with peer-reviewed and professional organization-supported standards. That allowed us to use a higher percentage of the data that we have, taking us from 1.3 million exams over a three-year period to 5.2 million exams over a one-year period.

KALPANA KANAL 11:41 One thing I want to point out is that we are using two completely different platforms. The paper from 2017 uses the American College of Radiology Dose Index Registry data, while the paper that we just published uses the Imalogix platform data, which is a lot more compact. I can’t directly compare the two because they’re totally different entities.

In defense of the ACR, when we published that paper, the Dose Index Registry was fairly new. We didn’t use multiphase data for that study, and it was a learning process for everybody. The ACR can only depend on what the sites send them; you don’t have control over what sites are doing. The same applies to Imalogix regarding what the site data is—if it’s garbage in, it’s garbage out. Curating that data took a long time when we did the work in 2017, which is why it took three years to publish the 2014 data.

This data is only good if you’re 100% sure that it is curated, good data with no unusual anomalies. Both papers have done their due diligence, but it definitely took us longer last time because of who was available to do the data analysis. Imalogix had that worked out so they could run the numbers and send an analysis in a day or two. It depends on your platform and who is working on it.

CHRIS ST. JOHN 13:40 That brings up an interesting point. Because all of this data is coming through the Imalogix platform, we have to address the fact that these exams come from facilities that have software in place that allows them to do deep dives on dose and image quality. They have a radiology optimization platform running. Does that potentially skew the data a little bit? What’s the story behind that?

KALPANA KANAL 14:20 I don’t think it skews the data. Anyone who has Imalogix at their facility is sending their dose data, and that’s what we used to publish the paper. It is the same with the ACR Dose Index Registry—anyone who subscribes is sending their data.

If you had Imalogix, you could do this analysis easily. But if you didn’t, you could still do it with any other dose optimization platform. If you don’t have a physicist or software, you can take a small chunk of your data and get a pretty good idea of what your median doses are going to be.

Using a specific platform does not mean this process cannot be replicated using other platforms or manually. It will just take longer manually to see what your numbers are.

THOMAS GRIGLOCK 15:17 Having been in academic medicine for almost two decades and then coming to work on a research endeavor with colleagues, I’ve come to understand that even with a vendor platform, we can still do really wonderful work together.

The initial feedback on the paper has been very positive. The commentary that accompanied it in Radiology was very positive and highlighted what we’re capable of doing.

Beyond being a research project, seeing a 22% reduction in your 75th percentile and a 9% reduction in your 50th percentile is really good news for a field that has gotten bad coverage for way too long. A lot of people are doing really strong work across the country, and this data quantifies that in a current manner. That is one of the bigger takeaways—we’re doing really well.

KALPANA KANAL 17:11 To add to that, Tom and I don’t want to give the message that we should continue to endlessly drop our doses. We should use the word optimization of doses. Low dose is not always a good thing, because the other side of this equation is image quality.

With the DIR paper from 10 years ago or this paper, we have no idea what image quality is being used at individual sites—they are just submitting dose data. While it is great that dose is reducing and moving in the right direction, I don’t want to send the message that we should keep dropping dose. We want to optimize our dose, keeping in mind that the image quality must remain diagnostic for our radiologists.

Everyone in the media is obsessed with reducing dose, but to me, that’s not the right phrase. It should be: let’s optimize our doses. Let’s not give more dose to our patients than necessary to get a diagnostic quality image.

THOMAS GRIGLOCK 18:26 Totally agree.

CHRIS ST. JOHN 18:26 You can see elements of that in the data. Chest CT with contrast dropped about 31%, whereas head CT without contrast only dropped about 3.5%. Do you mind explaining why that makes sense to you?

KALPANA KANAL 18:58 It makes sense because the head doesn’t change much after the age of six. Whether in the 2017 paper or 2026, head size doesn’t vary much whether you’re a 7-year-old child or a 70-year-old adult. Beyond six, the parameters and size don’t change much, so I would not expect to see a lot of dose reduction unless you are really dialing down the front end because of post-processing, like AI or other reconstruction techniques.

Head scans are complex even though it’s a small anatomy. Neuroradiologists really want to see that lovely distinction between gray and white matter, which is difficult to get at lower doses. So I’m not surprised we didn’t see a big reduction in head dose. Something technologically would have to be impactful enough to allow lower doses there.

Body sizes, on the other hand, vary a lot. You will see reductions there depending on calculation tools, like truncation correction. Chest CTs involve lungs and soft tissues where body sizes vary significantly, and we use much more AI-driven reconstruction algorithms in body exams than we do for the head.

CHRIS ST. JOHN 20:52 Do you know why that is?

KALPANA KANAL 20:56 It depends on the physicians. At our site, that gray-white distinction in the brain has to be great for them to make a diagnosis, and they haven’t warmed up to AI reconstruction algorithms on our scanners as much as the body group has. Because of variations in body size, data cleanup helps a lot in reducing dose on the front end in body exams. Tom, does that make sense to you?

THOMAS GRIGLOCK 21:30 Yeah. Doses represent the number of photons that make it out of a machine and reach the detector. The skull is the anatomical feature that keeps head doses from going down too much. You still need a certain number of photons to hit your detector to get a good image of the head—you can’t reconstruct your way out of that. In the chest and abdomen, you don’t have those same anatomical features messing up your X-ray statistics to the same degree, so you can do more in those areas.

KALPANA KANAL 22:34 What Tom is saying is that in the head, you have thick bone to penetrate, while the body is mainly soft tissue. That is why skull attenuation is such a key factor.

CHRIS ST. JOHN 23:01 Because of the skull and the biology of the head, we understand the doses that are necessary with current technology, so optimization in head scans is not as much of a factor because we’re already close to the limit. Is that fair?

THOMAS GRIGLOCK 23:38 There’s less leeway. In my experience, if radiologists get a suboptimal image in a chest or abdomen, unless it’s very bad, they’ll say, “It’s not how I like it, but I can deal with it.” Whereas with a head CT, it’s red or green—there’s no gradation. It’s either fine or it’s not.

KALPANA KANAL 24:14 In the head, they are looking for bleeds, brain perfusion, and stroke. It has to be good image quality to verify it’s an actual bleed and not an artifact. There isn’t much leeway in the brain.

Our data shows that we are pretty much optimized for the brain. I think our median for head CT without contrast was around 50 mGy, which is nearly identical to what we had in our 2017 paper. With the technology we have and what radiologists need, I’m comfortable saying we might be optimized for the brain unless something radical happens in technology.

CHRIS ST. JOHN 25:20 Your original paper has been out for about ten years now. Have you ever come across someone responding to their doses being too high in a knee-jerk or irresponsible way? What would be a silly reaction to having doses that don’t line up versus a measured and intelligent response?

KALPANA KANAL 25:56 That 2017 paper was very popular and got a lot of press attention because it was through the ACR and was the first time in this country that we had established national DRLs for the top 10 CT exams. I received many emails asking about our methodology.

An intelligent approach would be: “Let me compare my doses to the paper now that I have baseline data from across the country to see how I perform.”

A silly reaction would be letting ego get in the way and saying, “So my doses are high, big deal,” without being sensible about why they are high.

To be honest, our head doses at my institution are higher than the DRL in the paper. But I have a reason: I worked with my neuroradiologists, and that gray-white matter distinction is critical. Whether you are at 50 mGy or 60 mGy is not a huge magnitude difference to worry about. Our new paper lists 49 to 55 mGy for brain CT without contrast, and I’m running at 60 mGy on some of my scanners. I am okay with that because I know the image quality my doctors need on those scanners wouldn’t be sufficient if I reduced the dose. You acknowledge you’re higher, but as long as you’ve done your due diligence to understand and justify why, that’s great.

Now, if I were at 100 mGy compared to a DRL of 55 mGy, that is a problem. But being at 60 mGy is not a big deal if you have a valid clinical justification.

The message we want to send with data from 5 million exams today, or 1.3 million exams from 10 years ago, is that if your doses are different, you need to ask why. If you can justify it, that’s okay, but you must have a good reason. Some sites have older scanners that lack dose reduction options, which you have to keep in mind.

CHRIS ST. JOHN 29:51 What would you say to someone at a facility with older scanners who isn’t running photon-counting CTs and is working with what they have? How should they approach these DRLs and internal optimization?

KALPANA KANAL 30:14 Some older scanners, while lacking dose reduction bells and whistles, also don’t have large detector arrays that cause scatter and cone-beam artifacts. An older scanner works well; it just has fewer modern options, like an older car versus a new car.

You can still monitor and set protocols on older scanners to yield an optimized output and diagnostic image quality. It might not be at 55 mGy, but if it’s at 60 mGy, that’s okay. In my experience reviewing for the ACR, older scanners sometimes have lower doses because smaller detector arrays create fewer cone-beam artifacts and scatter.

THOMAS GRIGLOCK 31:42 By and large, newer scanners generally yield somewhat lower doses, but that doesn’t mean a newer scanner is always lower. We have other research projects showing that older machines can be driven appropriately.

Regardless of facility or equipment, seeing current national practice across a large sample gives context to people working in or consulting for hospitals. The equipment you have is what you have. Getting external context gives you insight into your own practices. Radiologists have different preferences, and patient populations vary.

This publication can drive conversations and provide insight into how facilities are performing and why.

KALPANA KANAL 34:32 Not every facility can afford dose monitoring software—it can be really expensive. Ten years ago, I used to tell people that you can subscribe to the ACR Dose Index Registry, which is much cheaper than dose monitoring software.

With the Dose Index Registry or this Imalogix paper, you are comparing your data against national peers. With internal software, you’re looking at doses within your facility. Comparing your performance to your peers is a good practice, but even without software, you can still monitor your doses in simpler, less expensive ways depending on your facility’s expertise.

CHRIS ST. JOHN 36:05 Was there anything in the dataset that surprised either of you?

THOMAS GRIGLOCK 36:21 Yes. The original goal of this project was updating the numbers, not necessarily doing a comparison. When we compiled the data from January 1st to November 15th of 2025 after that RSNA dinner, I pulled up Kalpana’s 2017 paper and put the 2017 and 2025 numbers side-by-side in Excel.

Seeing that clear decrease was the beginning of the comparison aspect of this paper, which became a huge story. In hindsight, these reductions make sense, but it wasn’t the initial objective. Realizing we had a much more compelling story than just a routine update was a wonderful byproduct of the process.

KALPANA KANAL 38:09 When we were having those conversations, we decided it didn’t make sense to present updated data without comparing it to what we had 10 years ago. Even if our top 10 protocols on Imalogix differed slightly, we needed to compare them to the original top 10 because anyone reviewing this would want to know how the data changed in 10 years.

Dr. Cynthia McCullough wrote an editorial for our paper referencing her earlier paper from 2006. If you look at the trajectory from 2006 to now, there has been a 35% reduction in dose overall. Everything tells a story. To know if you’re making progress, you have to compare your current state to where you were.

CHRIS ST. JOHN 40:05 Are there any other key points you think are worthwhile to cover before we wrap up?

KALPANA KANAL 40:26 We covered the methodology, the comparison, and the focus on optimization rather than just dose reduction. You might want to ask about what’s coming down the pike, specifically regarding the pediatric DRL paper.

CHRIS ST. JOHN 40:53 This updated DRL paper focused on adult exams, but what about pediatrics and other future studies? Tom, do you mind talking about future research coming out?

THOMAS GRIGLOCK 41:23 We have an update to the pediatric paper—which was previously published in 2021—that is currently under review. We are also close to submitting something on cardiac achievable doses and DRLs, which will be a first of its kind.

Pediatrics and cardiac imaging involve distinct nuances. Pediatrics is different because patient sizes range from tiny infants to adult-sized adolescents. Cardiac imaging is unique due to the technical nuances of capturing a motion-free snapshot of the heart at specific points in the cardiac cycle.

Finally, we are looking at the extent of imaging variability for these protocols across the 2025 data, examining it from both a dose and an image quality perspective. While previous papers include an asterisk stating that image quality is important without directly measuring it, our platform captures associated image quality metrics. That study will allow us to evaluate both dose and image quality side-by-side.

CHRIS ST. JOHN 43:50 If our listeners want to hear more about images being classified as “pretty good” or “good enough,” I encourage you to listen to our recent episodes with Dr. Sumei, where that is a central topic.

Dr. Kalpana Kanal and Dr. Thomas Griglock, thank you both so much for joining us today.

THOMAS GRIGLOCK 44:18 Thank you very much, Chris. This was great.

KALPANA KANAL 44:25 Perfect. I think that went well.

Questions from This Episode

What is the difference between a Diagnostic Reference Level and an Achievable Dose?
A Diagnostic Reference Level represents the 75th percentile of a dose metric across national data and serves as an investigation threshold, not a pass or fail limit. An Achievable Dose represents the 50th percentile and reflects a level facilities can realistically achieve under standard operating conditions. Facility protocol doses should generally fall between those two values.
Why did head CT doses decrease far less than body CT doses over the last decade?
Head CT doses dropped roughly 3.5% compared with a 31% drop in contrast-enhanced chest CT because the skull attenuates x-rays and requires a stable photon threshold to preserve diagnostic image quality. Neuroradiologists need clear gray-white matter differentiation, leaving less room for aggressive dose reduction.
How can a facility compare its protocols against national DRLs without expensive software?
Facilities without dedicated dose management software can review about 50 recent exams for a specific protocol, calculate median dose metrics, and compare that internal median with published national DRL percentiles. The comparison helps medical physicists and clinical leaders identify protocols that need further review and optimization.