This paper lands like a last-second three that sends the arena into chaos, except the replay shows the hero shot may have brushed the rim, the backboard, and possibly the diagnostic rulebook on the way in. Researchers looked at MRI scans from more than 11,000 adults in Germany and found sacroiliac joint bone marrow oedema in about 30% of the general population - roughly 50 times the prevalence of self-reported axial spondyloarthritis in the same cohort [1]. That is not a small correction. That is the kind of result that makes radiologists, rheumatologists, and anyone who has ever stared nervously at an MRI report sit up a little straighter.

The basic issue is this: bone marrow oedema in the sacroiliac joints has become one of the key MRI clues for diagnosing axial spondyloarthritis, an inflammatory disease that can cause chronic back pain and long-term joint damage. MRI helped clinicians catch disease earlier, which is good. Genuinely good. But this study asks a sharper question: what if the "classic" MRI sign also shows up all over the place in people who do not have inflammatory disease?

The suspiciously popular MRI finding

Bressem and colleagues analyzed participants aged 20-69 from the German National Cohort. A subset of scans was read by masked experts, and the rest were assessed with a validated deep-learning system that segmented and quantified oedema volume automatically [1]. Both approaches landed in the same neighborhood: about 29% to 31% prevalence.

That is the headline. The second punch lands right after it: the oedema tracked more strongly with pregnancy history, higher body weight, occupational physical strain, and recreational physical activity than with inflammatory disease itself [1]. In women, previous pregnancy was associated with higher odds of oedema. In men, age and intensive physical activity stood out. If you were hoping MRI alone would act like a tidy lie detector, the machine has chosen chaos.

This fits with a growing body of work showing that sacroiliac MRI changes are not exclusive to axial spondyloarthritis. Pregnancy-related studies have shown postpartum oedema can look a lot like inflammatory sacroiliitis on imaging, especially early after delivery [2,3]. Recent reviews now explicitly warn about mechanical and postpartum "mimics" when interpreting these scans [4,5].

Why this matters more than it sounds

This is not some technical squabble between people who enjoy arguing about grayscale blobs for sport. A positive MRI can influence whether someone gets labeled with a chronic inflammatory disease and whether they start potent anti-inflammatory drugs. Those treatments can help the right patient a great deal. They are not a casual multivitamin.

That is why the paper matters. It does not say MRI is useless. It says MRI findings need context, the way a smoke alarm needs context. Burnt toast and a house fire can both make the thing scream. You do not evacuate the neighborhood because someone got ambitious with breakfast.

The timing is interesting, too. In the past few years, specialists have been getting louder about an "imaging crisis" in axial spondyloarthritis: MRI made earlier detection possible, but bone marrow oedema by itself has weak specificity and can invite overdiagnosis [5]. A 2023 analysis also found sex-specific differences in MRI diagnostic performance, which is another way of saying the "one-size-fits-all" approach was doing what one-size-fits-all usually does: fitting some people badly [6].

The AI angle: useful, impressive, and exactly why we should be careful

There is also an AI subplot here, and it is worth noticing. The study used deep learning to scale MRI assessment from a manually read sample to a national cohort [1]. That is a real capability jump. Without automation, this sort of population reference map would be painfully slow and probably expensive enough to make grant reviewers start blinking.

But the result also carries a quiet warning for medical AI. If you train a system to detect a finding that is common but nonspecific, congratulations: you may have built a very efficient way to industrialize ambiguity. The model is not "wrong" if it finds oedema. The danger comes when humans treat that output as a diagnosis rather than one clue among several. The capability gains are genuinely impressive. And that is precisely why the judgment layer matters more, not less.

Recent imaging guidance and review work point in the same direction: combine MRI with clinical history, symptoms, lab data, lesion pattern, and structural findings rather than worshipping a bright spot on a STIR sequence like it descended from the mountain carrying stone tablets [4-7].

The bigger takeaway

The most valuable thing in this paper may be its ordinariness. It reminds us that bodies are messy, load-bearing, history-collecting machines. Pregnancy leaves traces. Weight-bearing leaves traces. Exercise leaves traces. Aging leaves traces. Not every biological footprint is a disease signature.

For clinicians, that means more careful interpretation. For researchers, it means better thresholds, better lesion definitions, and probably better multimodal models. For patients, it means an MRI is part of the conversation, not the entire verdict.

And for the rest of us, it is a useful correction to the usual "AI plus imaging equals certainty" storyline. Sometimes the smartest system in the room still needs someone to say, calmly, "let's check whether this is inflammation or just the skeleton filing a complaint about physics."

References

1. Bressem K, Torgutalp M, Diekhoff T, et al. Prevalence and determinants of sacroiliac joint bone marrow oedema in the general population in Germany: a population-based cross-sectional study. Lancet Rheumatology. 2026. DOI: 10.1016/S2665-9913(26)00071-8. PubMed: 42105782

2. Renson T, Depicker A, De Craemer AS, et al. Evolution of Magnetic Resonance Imaging Lesions at the Sacroiliac Joints During and After Pregnancy by Serial Magnetic Resonance Imaging From Gestational Week Twenty to Twelve Months Postpartum. Arthritis Rheumatol. 2023. PubMed: 36704824

3. Hoballah A, Lukas C, Leplat C, et al. MRI of sacroiliac joints for the diagnosis of axial SpA: prevalence of inflammatory and structural lesions in nulliparous, early postpartum and late postpartum women. Ann Rheum Dis. 2020. PubMed: 32522743

4. Diekhoff T, Hermann KGA, Poddubnyy D, et al. When it is not sacroiliitis. Skeletal Radiology. 2025. PubMed: 40461872

5. Diekhoff T, Poddubnyy D. The imaging crisis in axial spondyloarthritis. Lancet Rheumatology. 2025;7(9):e652-e656. DOI: 10.1016/S2665-9913(25)00108-0

6. Ulas ST, Proft F, Diekhoff T, et al. Sex-specific diagnostic efficacy of MRI in axial spondyloarthritis: challenging the "One Size Fits All" notion. RMD Open. 2023;9(4):e003252. DOI: 10.1136/rmdopen-2023-003252. PMCID: PMC10619004

7. Malaviya AP, Kucybała I, Lecler A, et al. Diagnosis, monitoring, and management of axial spondyloarthritis. Rheumatology International. 2024. DOI: 10.1007/s00296-024-05615-3

Disclaimer: This blog post is a simplified summary of published research for educational purposes. The accompanying illustration is artistic and does not depict actual model architectures, data, or experimental results. Always refer to the original paper for technical details.