The GLP-1 drug class has delivered remarkable results across cardiometabolic disease, kidney function, and cardiovascular outcomes. Its effects in knee osteoarthritis are increasingly recognized. But according to Professor Mikael Boesen, Professor of Musculoskeletal Radiology and Head of Imaging Research at Department of Radiology, Copenhagen University Hospital Bispebjerg and Frederiksberg in the Capital Region of Denmark, the field has a measurement problem. Most large GLP-1 Knee Osteoarthritis (KOA) trials are still enrolling patients based on a standing plain radiograph of the knee in one projection and the Kellgren Lawrence (KL) Grading scale from 0–4 published in 1957. This images only the bone in 2D, while the tissues where this drug likely acts — synovium, cartilage, muscle, bone marrow, fat pads — go largely unmeasured. Professor Boesen has spent most of his career building the evidentiary case for advanced imaging of musculoskeletal diseases using MRI, ultrasound, CT, DEXA, and quantitative imaging in musculoskeletal diseases and clinical trials. In this conversation, he explains what the current imaging toolkit can and cannot tell sponsors, where endpoint design is going wrong, and which technologies are underused relative to the evidence supporting them.
How do the main imaging modalities differ in what they can actually detect in a knee joint?
First and foremost, we need to understand how GLP-1 drugs work on the tissue of interest in the joint — and whether they work equally across all types of patients. That is the big question, and it’s still one we cannot fully answer.
Here one reason is that most of the large GLP-1 trials have only used the plain radiographs as the inclusion criterion without adding more imaging at baseline or follow-ups. Radiographs are the oldest, most widespread, robust and validated imaging modality we have for Knee OA imaging, but it has real limitations for detecting what we actually want to know, as it only visualizes the bones in two dimensions and the space between them as an indirect measure of the cartilage and supporting tissues like the menisci.
Since we still primarily are using the KL Grading of 0 to 4 in 21st-century trials, where 0 is normal, 1 is pre- and 2 involve early KOA with established osteophyte formation, and 4 is late-stage disease where patients typically proceed to joint surgery. We do not know how these drugs actually work on the soft tissues — the synovium, the inflammation, cartilage, bone and muscles. Most trials simply do not include imaging that allows to measure relevant changes in those structures.
The reason radiographs remains dominant is that it has long been the FDA reference standard. To get approval for a disease-modifying OA drug, you have to show on radiograph that joint space narrowing has slowed or stabilized. That is extremely difficult. You need large patient numbers and very long follow-up to see differences on radiographs — which is part of why we are still in a phase where we do not truly understand the drug’s mechanism of action in this tissue.
That said, the few trials that have gone beyond radiographs and used MRI, ultrasound, CT or DEXA alongside it have shown reduction in inflammation in patients who lose weight. But the association between weight loss and inflammation in OA is not very strong overall, and I think that it is partly because radiographic-based enrollment is mixing together very different patient types.
Over the last five to six years, the concept of OA phenotypes has become more and more important. Within that broad pool of patients diagnosed with osteoarthritis, there are subgroups driven by very different mechanisms — inflammatory, post-traumatic, crystal-driven, aging-related and genetic, and metabolic. The metabolic phenotype is the one associated with obesity, insulin resistance, hypertension, and type 2 diabetes — what we know as metabolic syndrome. That is in theory the subgroup most likely to respond to a GLP-1 drug, and possibly even within that, there may be further sub-phenotypes where the effect is even stronger.
There is an ongoing trial right now that I find quite interesting — a GLP-1 drug delivered directly into the joint, led by Francis Berenbaum, using a broad imaging package including advanced MRI sequences to try to understand the mechanism in a randomized controlled design. The outcome of that trial will be very informative.
So there is no clean answer yet, but the interest is high and the framework is getting sharper. We really do need to understand how the drug’s mechanism of action is reflected in the imaging signal before we can design trials that reliably detect it.
“Within that large pool of patients, there are subgroups with osteoarthritis for various reasons. The phenotypes have become more and more interesting to understand — because there may be subgroups that actually benefit, and possibly even sub-conditions within those.”
How should that understanding of tissue-level biology drive endpoint selection in OA trial design?
Professor Mikael Boesen
Professor of Musculoskeletal Radiology & Head of Imaging Research, Copenhagen University Hospital
When I am asked that, my first question back is always: what is your mechanism of action? What do you know from early-stage work — from animal studies, from phase 1 toxicology? Because the answer really depends on which tissue the drug is targeting and which phase you are in.
In the earlier phases — phase 1 and 2 — you have the best opportunity to use more advanced imaging to get genuine insight into how your drug behaves in tissue. If your drug is highly active at the synovial receptor level, then the trial design should obviously be built to capture synovial behavior. That means selecting patients who have active synovitis, and using ultrasound or MRI to confirm that at inclusion. You are looking for a phenotype that has the inflammation your drug is targeting.
If the drug impacts bone, that is a different design — you are thinking about broader structural metrics focusing on the bone marrow with MRI or even 3D bone models from either MRI or CT scans. If the primary target is cartilage, then you need 3D cartilage sequences that can be segmented to track volume and thickness over time, and ideally sequences that also capture what is happening inside the cartilage, not just its morphology.
The most available of those compositional sequences right now is T2 mapping, which reflects water content and free water in tissue. When cartilage begins to degrade, it allows more free water into the matrix of collagen and proteoglycans before it collapses and reduces in thickness, and you can see that in the T2 map. There are also more research-driven sequences like T1rho and dGEMRIC, but these are more complicated to perform and not standard sequences in most MRI scanners. In general I would recommend using the more advanced sequences and imaging procedures in the earlier phases of the drug trials to build understanding of the drug behavior on various tissues in different patient phenotypes, that can influence and guide the design of later stage when moving toward the pivotal phase 3 trial.
The sensitivity of your imaging tool also has a direct impact on sample size. A more sensitive, targeted measure of the specific biology you are after allows you to work with a smaller sample. A crude measure requires a much larger one. So the imaging decision is not just scientific — it has real implications for trial feasibility and budget.
All of this comes down to a very early dialogue. What is the drug doing? Which patient population are you studying? Can you actually measure the endpoint you want? There is no universal recipe. It really depends on that specific combination of drug, tissue target, and patient phenotype.
How do you run a rigorous multi-site DEXA program, and where do these studies most commonly go wrong?
DEXA is a well-established, 30-to-40-year-old technology, and it has been continuously refined. It gives you information about bone composition, and it allows you to look up a reference normative diagram and compare a patient’s values against age- and sex-matched populations — the T-score and Z-score. The standard measurement sites are the neck of the femur of the hip, the spine from L1 to L4, and sometimes the radius.
The challenge in multi-site trials is ensuring that measurements are comparable across scanners. Different machines produce different values, even for the same patient, and that is where standardization becomes critical. The solution is a standardized phantom — a physical calibration object that circulates between sites so each scanner can be cross-calibrated. If you do not think carefully about standardization from the beginning, you end up with incomparable data from different scanners with different software versions. That is a failure mode I have seen.
Beyond bone, DEXA can also measure lean body mass, which has become increasingly relevant given the questions around muscle health and GLP-1 drugs. There is evidence that the weight loss induced by this class of drugs includes loss of lean mass, and there is ongoing debate about why that is happening. But the limitation of DEXA for this purpose is that lean body mass is not a true muscle measure. It captures all connective tissue between fat and bone — tendons and interconnective tissue structures — not just muscle itself. So if you want to understand specifically what is happening to muscle, DEXA is not sufficient on its own.
The GLP-1 drugs seem to have positive effects that reach into a lot of systems — improved cardiometabolic disease, kidney disease, cardiovascular outcomes, and reduction of KOA symptoms. But what we do not yet fully understand is what is happening to muscle and bone. The latest real-life evidence of GLP-1 drugs — especially the more potent dual-effect drugs — show a small but statistically significant reduction in bone density in the treatment arm relative to placebo or other single-effect GLP-1 drugs. Why that is happening is still being worked out. It could be a molecular-level effect on muscle and bone remodeling balance, or it could partly be nutritional — patients are eating less and may not be getting sufficient vitamin D and other nutrients needed to maintain healthy bone. For muscle and lean body mass, the evidence shows overall that the weight loss achieved is also affecting the lean body and muscle mass, which accounts for up to 30% in some studies. The reason for this could also be nutritional combined with insufficient physical activity to maintain muscle mass while losing weight. There is no strong answer yet, although recent results point towards that muscle mass and physical function can be improved in patients that actively take part in physical exercise programs.
If you want to understand what is happening to muscle, you really do need MRI, because that is the reference standard for separating fat from muscle tissue. In my opinion, you cannot discuss muscle health in isolation from bone health with these potent new weight-loss drugs. Even if your drug’s effect targets one of them specifically, the two systems are interconnected — especially in persons with metabolic syndrome — which is important to understand by using specifically designed trials that allow you to measure the effects on both tissues.
Why does imaging-based muscle assessment matter in GLP-1 trials, and what is myosteatosis?
Myosteatosis is essentially fat infiltration of muscle. It happens as we age. We have done a great deal of research on this, and the picture that emerges is fairly clear: as a rule of thumb you can build muscle up until around age 60. After 60, you can maintain muscle mass through consistent resistance exercise, but you cannot add much more, at least not at any meaningful pace. After 70, it is very difficult to build new or even maintain muscle mass — you can only try to slow the rate of loss. Exercise is the solution to this and still helps regardless of age, but the best results come if you have built enough muscle at an earlier age to prevent early sarcopenia effects in older age.
Myosteatosis is the process by which fat infiltrates the muscle fibers over time, and the end state of that process is sarcopenia — where the muscles are weak enough that you begin losing the ability to do ordinary things: getting out of a chair, climbing stairs. The message is that muscle has to be built while you still can, so that you have a higher baseline to lose from.
MRI is the tool that lets you properly distinguish fat from functional muscle tissue, which is something DEXA cannot do with the same precision. DEXA captures lean body mass, but that includes connective tissue and tendons — not just muscle fibers. If you want to understand whether a drug is protecting or degrading actual muscle quality, MRI gives you that separation in a way DEXA cannot.
In phase 3 trials, DEXA is still the standard for bone endpoints. But MRI is becoming more and more important for understanding what is happening in muscle, and in certain disease areas — like myositis — MRI is already the mandatory choice for demonstrating that a drug prevents further muscle deterioration or reduces muscle inflammation.
There is also an interesting discussion here around CT. CT scans capture bone as a natural by-product of scanning for other indications, and they are three-dimensional, which gives them advantages over the two-dimensional projections DEXA provides that struggle with consistent analysis in patients with large weight loss. Opportunistic screening on existing CT scans is becoming increasingly recognized as a way to detect bone loss earlier and potentially prevent fractures before they happen. In my view, that will become more standard over time, though it is not quite there yet.
Why are dynamic contrast-enhanced MRI and quantitative CT still underused in musculoskeletal trials, given the evidence for them?
It depends a lot on availability, because I do see a lot of exciting technologies where the latest generation scanners called Photon Counting CT can visualize structures below 100 microns — things you simply could not see before across all disease areas. The potential there is significant: bone behavior at a trabecular level rather than the cruder density measure you get from DEXA, early cancer detection, cardiac disease. But these systems are still not widely available enough to rely on across a multi-site trial.
Even before you get to those emerging tools, we have not fully utilized what the current generation already offers. Dual-energy CT — what we call spectral CT — has the capability to provide material decomposition, which means you can characterize tissue composition in ways that go beyond conventional CT. That capability is still underused relative to what the technology can actually deliver.
The same is true of Dynamic Contrast-Enhanced MRI. In my view, this is underutilized across musculoskeletal and inflammatory joint diseases. The kinetics of the contrast agent reflect synovial inflammation and tissue inflammation more accurately than static sequences post contrast imaging, and we have robust evidence from rheumatoid arthritis as well as knee OA that perfusion measured this way is the best imaging correlate of histologically graded tissue inflammation, and that better associates with pain than any other imaging method. It is well established in prostate cancer, breast cancer, and other oncology indications. In musculoskeletal disease it tends to be perceived as technically complicated or cumbersome — which it really is not. But that perception has slowed adoption.
It is a sequence that all available scanners on the market can do, it takes 3–4 minutes more scan time and has robust post-processing software available. I believe that DCE-MRI should be considered in all imaging trials where you have already decided to give intravenous contrast to obtain static post-contrast images.
On the AI side, we are beginning to see automated segmentation tools that can handle things that used to require extensive manual work — drawing cartilage volume in all submillimeter slices of a knee in all joint surfaces or on every muscle across 35-plus centimeters of thigh on both sides, for instance. That is the kind of work that would dominate a trial budget if done manually, and automation is starting to make it genuinely feasible.
The driver that moves all of this into broader use is, in my view, clinical trials. Controlled trials in a relatively small and controlled sample can demonstrate the value of a more sensitive imaging measure in a way that you can then later transfer into clinical practice. Oncology has led the way on this because the stakes are high and acute — these are life-and-death decisions on a short timeframe. Musculoskeletal disease moves more slowly as a field, even though obesity, sarcopenia, and OA represent some of the largest threats to quality of life and costs for society due to sick leave, loss of working hours and premature retirement pensions in the developed world today, expected to exponentially increase in the near future. But on the good side there are very exciting new drug molecules in development in this space, and I think the imaging infrastructure to study them properly is getting closer to where it needs to be.
This content is sponsored by Image Analysis Group.
Moe Alsumidaie, MBA, MSF, is founder and Chief Editor of Vanguard Publications, which publishes Clinical Trial Vanguard, Pharma Vanguard and BullScope, and Head of Research at CliniBiz. He has two decades in clinical trial operations and data science, with earlier roles at Genentech, Abbott Vascular and Stanford University Medical Center, and is a guest lecturer in clinical trial sciences at Rutgers University.
