Investigating the associations between lumbar paraspinal muscle health and age, BMI, sex, physical activity and back pain using an automated computer-vision model : a UK biobank study

Background context The role of lumbar paraspinal muscle health in back pain (BP) is not straightforward. Challenges in this field have included the lack of tools and large, heterogenous datasets to interrogate the association between muscle health and BP. Computer-vision models have been transformative in this space, enabling the automated quantification of muscle health and the processing of large datasets.

Purpose To investigate the associations between lumbar paraspinal muscle health and age, sex, BMI, physical activity, and BP in a large, heterogenous dataset using an automated computer-vision model.

Design Cross-sectional study.

Patient sample Participants from the UK Biobank with abdominal Dixon fat-water MRI (N=9,564) were included (41.8% women, mean [SD] age: 63.5 [7.6] years, BMI: 26.4 [4.1] kg/m2) of whom 6,953 reported no pain, 930 acute BP, and 1,681 chronic BP.

Outcome measures Intramuscular fat (IMF) and average cross-sectional area (aCSA) were automatically derived using a computer-vision model for the left and right lumbar multifidus (LM), erector spinae (ES), and psoas major (PM) from the L1 to L5 vertebral levels.

Methods Two-tailed partial Pearson correlations were generated for each muscle to assess the relationships between the muscle measures (IMF and aCSA) and age (controlling for BMI, sex, and physical activity), BMI (controlling for age, sex, and physical activity), and physical activity (controlling for age, sex, and BMI). One-way ANCOVA was used to identify sex differences in IMF and aCSA for each muscle while controlling for age, BMI, and physical activity. Similarly, one-way ANCOVA was used to identify between-group differences (no pain, acute BP, and chronic BP) for each muscle and along the superior-inferior expanse of the lumbar spine while controlling for age, BMI, sex, and physical activity (α=0.05).

Results Females had higher IMF (LM mean difference [MD]=11.1%, ES MD=10.2%, PM MD=0.3%, p<.001) and lower aCSA (LM MD=47.6 mm2, ES MD=350.0 mm2, PM MD=321.5 mm2, p<.001) for all muscles. Higher age was associated with higher IMF and lower aCSA for all muscles (r≥0.232, p<.001) except for LM and aCSA (r≤0.013, p≥.267). Higher BMI was associated with higher IMF and aCSA for all muscles (r≥0.174, p<.001). Higher physical activity was associated with lower IMF and higher aCSA for all muscles (r≥0.036, p≤.002) except for LM and aCSA (r≤0.010, p≥.405). People with chronic BP had higher IMF and lower aCSA than people with no pain (IMF MD≤1.6%, aCSA MD≤27.4 mm2, p<.001) and higher IMF compared to acute BP (IMF MD≤1.1%, p≤.044). The differences between people with BP and people with no pain were not spatially localized to the inferior lumbar levels but broadly distributed across the lumbar spine.

Conclusions Paraspinal muscle health is associated with age, BMI, sex, and physical activity with the exception of the association between LM aCSA and age and physical activity. People with BP (chronic>acute) have higher IMF and lower aCSA than people reporting no pain. The differences were not localized but broadly distributed across the lumbar spine. When interpreting measures of paraspinal muscle health in the research or clinical setting, the associations with age, BMI, sex, and physical activity should be considered.

Mots clés

adipose tissue; artificial intelligence; back muscles; fatty infiltration; low back pain; magnetic resonance imaging