The patient in front of you has had chronic low back pain for four years. Their MRI shows paraspinal muscle wasting — fat infiltration visible in multifidus, atrophy of erector spinae. You can see it. The question is what to do with it. Whether the imaging tells you anything actionable about how disabled this patient is and where rehabilitation should direct its effort.
A prospective study published in European Radiology in December 2025 asked two related questions: can deep learning accurately quantify that muscle fat fraction from standard MRI? And what is the actual relationship between that measurement and functional disability?
The technical finding
The study recruited 96 patients with chronic low back pain and 86 healthy controls. All underwent 3T MRI. A deep learning model quantified multifidus and erector spinae fat fraction from 3D T2-weighted images, validated against Dixon MRI — the gold standard for muscle fat quantification. Concordance correlation coefficients were 0.96 for multifidus and 0.95 for erector spinae.
That is a technically useful result on its own. Dixon imaging is not universally available and is not routinely acquired in most CLBP MRI protocols. A DL tool extracting equivalent information from a standard T2 sequence removes a barrier to routine muscle composition assessment without requiring additional acquisition time or scanner configuration.
The clinical finding
The more important result was in the mediation analysis.
Paraspinal fat fraction correlated significantly with disability scores — Oswestry Disability Index and Roland-Morris Disability Questionnaire, r = 0.25–0.49. Muscle endurance correlated more strongly: r = −0.57 with ODI. Mediation analysis showed that 27–100% of the relationship between fat fraction and disability was indirect — operating through muscle endurance, not through structural degeneration directly.
The fat infiltration is not directly disabling the patient. It is disabling them substantially through the functional loss it produces. The muscle has degenerated, the degeneration impairs endurance, and impaired endurance is what drives the disability score.
Why this matters for treatment
Structural degeneration is largely fixed on a clinical timeline. Muscle endurance is not.
If the pathway from structural change to disability runs substantially through function, a rehabilitation programme targeting endurance has a mechanistic basis for improving disability outcomes even in a patient whose MRI looks bad. You are not fixing the anatomy. You are interrupting the functional consequence of it. That is a clinically meaningful distinction — it shifts the rehabilitation question from “what can we do given the imaging?” to “what functional capacity can we restore regardless of the imaging?”
Limitations
Ninety-six CLBP patients from a single Chinese military hospital is not a representative sample of a UK spine clinic population. The mediation model is cross-sectional — it does not prove that improving endurance will reduce disability, only that the two are associated through a plausible pathway. Prospective intervention data are needed to confirm direction of effect. The DL tool itself requires external validation before clinical adoption.
The practical implication
In patients with visible paraspinal muscle wasting on MRI, the imaging is not just a pathological description — it is a prompt to assess functional capacity. Muscle endurance testing is not routine in outpatient spine practice. This study gives a reason to consider it.
If the degeneration-to-disability relationship runs through endurance, that is where rehabilitation should direct its effort. Not the anatomy. The function the anatomy is producing.
References
- Chen P, et al. Deep learning-based assessment of paraspinal muscle degeneration and its relationships to muscle function and disability outcomes in chronic low back pain: a prospective study. Eur Radiol. 2025. PMID 41405693. https://doi.org/10.1007/s00330-025-12171-2