While CT protocols can provide a template for how to image certain parts of the body, they can leave a lot to be desired in terms of optimizing the scan for the individual patient being examined. The indication for the exam, the size and gender of the patient and the necessary image quality for an accurate diagnosis are all important variables that don’t always get consideration.

In order to allow for consideration of these variables and others, Dr. Donald P. Frush and his colleagues at Duke Medical Center have devised a quantitative method that simplifies the personalization of CT scans. HealthCare Business News spoke to him about what this method brings to the table, how AI could eventually raise the bar on what they’re already accomplishing and the vital role of manufacturers in bringing these capabilities to imaging facilities.

HCB News: Why is consistency important in the way pediatric patients receive CT scans?

Dr. Donald P. Frush: If we talk about CT scans, one of those risks is radiation. One other would be contrast administration, but oftentimes the risk profile of CT is focused on radiation. There is some variability in both of those things. For example, some people may tolerate or be able to use a little bit lower image quality and render sufficient diagnoses. Some people may need higher image quality, thus the radiation dose would vary with each of those.

There’s not a specific amount or dose or particular image quality. I think when it comes to consistency you are informed about what you need to do in your practice. You are informed about the kind of image quality you need and you have adjusted your CT scanning to provide that level of image quality, minimizing the amount of radiation that’s there. It’s sort of adhering to that.

That consistency should be for all your protocols. Over time, you shouldn’t do things differently. That may vary in a small practice in Wyoming versus a children’s hospital in Philadelphia versus a university-based program in Dallas.

It’s accountability and what you do and how you define what you need and what risks you minimize in getting there. It’s informed best practice and maintained throughout your CT practice and over time. That’s kind of what consistency is. Sometimes, when people say you should be consistent with X, Y and Z, everyone assumes they should do the same thing. It’s not quite that simple.

HCB News: You and your colleagues have devised a quantitative method for ensuring quality scans with proper radiation dose. What factors are involved in the use of this method?

DF: It’s one example of how this can be done to optimize your practice. It doesn’t say because this was published or what we did, now everyone can know how to do a CT scan in a child. What that model does is take into account things that we need to think about when we do CT scanning. That starts first of all with the questions, ‘What are you looking for? What kind of disease process or disorder are you looking for?’

It’s not necessarily chest CT-based or abdomen CT-based, meaning you don’t just do an abdomen CT for everything. Depending on what you’re looking for, that abdomen CT would be modified. It’s clinical indication and it’s size. That size from a public standpoint equates to age, which is not quite correct because you can have big young people and small old people. The size that we talk about is really cross-sectional area CT. It’s sometimes body surface area. It’s sometimes age. Those are variable, but the general term is size. The best physical dimension is the thickness of the patient.

Aside from indication and size, what is the gender? What organ are you looking at when you’re trying to calculate and estimate risk? Girls, for instance, might have different sensitivities to radiation due to breast tissue and some other considerations, making this a gender-related matter.

It’s also what kind of image quality you want. Someone may feel comfortable looking at a noisy image for kidney stones and someone else may not feel that they can do that. It’s what the radiologist needs in overall image quality. It’s the scan indication. It’s the size of the patient and gender considerations related to risk.

What we did was take all of those ingredients and put them together. If you adjust doses, your risk is going to change. Your image quality is going to change. It’s the interplay in the relationship of those so that, in the end, taking all of those things into account for a specific indication, which might be lung nodules in a pediatric patient, you come out with a protocol that is accountable to all of these things.

HCB News: How does this approach to imaging compare with the conventional approach at most facilities?

DF: I think that most imaging professionals -- radiologists, technologists, medical physicists, that sort of a group of imaging professionals -- have an understanding and respect for these kinds of things. People understand that we use radiation, that radiation may cause injury. People understand we may need to do things a little differently based on size, and people have a general understanding that some types of exams of the abdomen can be done a little differently than other types of exams of the abdomen.

The protocols are not always mindful of all of those things. Generally, I think protocols take into account indication, so if you’re looking for a small liver lesion, you would do it differently than looking for a kidney stone, than looking for a valve obstruction, than doing an evaluation of the lumbar spine, which would also necessitate an abdomen scan. I think those are different protocols or set-ups and people understand that.

I would say that oftentimes, that if you have a 110-pound young man versus a 250-pound individual, sometimes the adjustments are not made or made appropriately to account for differences in size in those individuals with respect to certain parameters. The machine does some of those things. We’ll look at it and say, ‘OK. This is a bigger person. We need to X versus Y.’

The first thing is different protocols for different indications, such as kidney stones versus kidney lesions versus something else. Then, it goes down the list on these other three things, and I think there is successfully less attention paid and less modifications done in protocols.

What we did was simply to say all of these elements are important and we created a model where all of these elements are simply blended together to provide a framework or a roadmap to set your scan depending on how you want to emphasize the image quality. It was sort of a little bit of a tool kit for the individual to not have to do the fine evaluation of image quality, to not have to look at radiation dose. We sort of provided those elements to craft a tool for the range of sizes for this desired image quality. This is what the settings would be.

To the best of my knowledge, there’s no protocol or widely available set of protocols that combine all those elements for you. There are variable contributions of maybe one, two, three elements, but those aren’t all put in there. It’s usually just manufacturer-based protocols with some modifications that university-based programs or people might do, but generally, a protocol for an abdomen CT, giving it a meaning or when equipment is put in, the manufacturer will say, ‘Well, these are protocols that this institution is using and people just sort of put these on and use them.’

It’s not to say that’s bad. Many places do quite good work, providing diagnostic information at doses that are very reasonable, but what our model does is say, ‘OK. What is everything that we’re accountable for, and how do we put all of these things together in a tool kit where we would work to get these protocols across many indications out there?’ It’s fine-tuning it a little bit and developing something where a private community practice doesn’t need to go through what was done to develop this. What were the various components, how they affected scan quality, and so on.

HCB News: Could this quantitative approach be implemented into scanning software some day?

DF: Absolutely! It takes some work. It would take more protocols to be done. It would take integration with manufacturers to build this either into their equipment, or to provide information that installation could be put on there. It would take some time to do that.

The first thing is that people need to accept this is what we need to do. People don’t like to change unless they have to, or unless there’s some reward or motivation. Then, the broad medical community says, ‘Hey, in being accountable for what we do, we have to have all these elements. It needs to be personalized based on the size of the patient. It needs to be personalized based on what the radiologist is looking for in image quality. It needs to be personalized in terms of what are the risks. Should a different technique be used for a different 15-year-old girl than a 15-year-old boy?’

All of those things need to come together because we are responsible for those things. It’s getting the community to accept that all of these elements are important and moving in that direction.

This is sort of an early work to say this can be done and once it’s embraced and accepted, then other places can begin to work on that concept of those components, all being important in protocol development.

HCB News: Is dose customization an area where deep learning algorithms are poised to change CT scans?

DF: Yes, absolutely. If we can take the human model, and if the machine can learn to do what humans do, then, for example, if a person looks at 100 CT scans of something and detects something in a certain period of time, the machine can look and say, ‘OK. They were much better when the scan was this way and they were much worse when the scan was this way and here are all the elements that contribute to the poorer performance and here are all the elements that contribute to better performance.’

HCB News: What needs to happen to move this approach to imaging forward?

DF: In order to do this right, it is going to require some teamwork and cooperation. People need to understand this is not just radiologists or pediatric radiologists looking at CT scans. This is a community that is increasingly invested in computational sciences, in deep learning, in IT, in medical physics and so on. It’s an expanded view. It’s intimately involved with the manufacturers. If manufacturers are not onboard, then changes will happen much slower. So, industry is a really important part of this.