A large-scale investigation using deep learning to evaluate abdominal computed tomography scans has found no meaningful average difference in bulk muscle quantity between individuals diagnosed with low back pain and matched control subjects. Led by researchers including Kathryn R. Marusich, the cross-sectional study demonstrated that bulk skeletal muscle cross-sectional area normalized to height squared did not significantly differ between the cohorts. The findings challenge assumptions regarding direct muscular volume deficits in back pain while confirming the feasibility of opportunistic automated body composition analysis using routine clinical imaging datasets.
The research team applied an artificial intelligence model to opportunistically quantify tissue cross-sectional areas from abdominal scans across a broad clinical population of 23,508 individuals. By analyzing routine diagnostic imaging, the investigators established normative body composition trends across age and body mass index, evaluating a normative analysis group of 10,835 subjects comprising 6,121 females. The automated measurements captured skeletal muscle area alongside visceral adipose tissue and subcutaneous adipose tissue, allowing researchers to evaluate anatomical patterns at a scale significantly exceeding traditional imaging investigations.
To specifically explore the relationship between body composition and musculoskeletal symptoms, the authors leveraged patient electronic health records to establish a clinical cohort of 2,622 individuals diagnosed with low back pain, including 1,623 female patients. They subsequently formed an age-, sex-, and body mass index-matched control cohort consisting of 2,577 subjects, of whom 1,594 were female. This matched configuration allowed the researchers to conduct comparisons within a study population twenty-five times larger than most previous imaging-based low back pain investigations.
The opportunistic automated measurements reproduced established physiological patterns across adulthood, including normative skeletal muscle loss rates of five to ten percent per decade alongside an increasing ratio of visceral to subcutaneous adipose tissue in older individuals. When comparing the clinical cohorts, the low back pain group exhibited a statistically significant reduction in the ratio of muscle to total abdominal adipose tissue compared to matched controls (p < 0.0001). However, the statistical effect size for this difference remained minimal, with an eta squared value of 0.004.
Beyond the small divergence in relative muscular-to-fat proportions, the low back pain and control cohorts demonstrated no differences in the ratio of visceral to subcutaneous adipose tissue or in height-normalized muscle area. These observations indicate that gross deficits in muscle mass are not a primary distinguishing feature of clinical low back pain. Overall, the agreement between automated findings and recognized physiological trajectories establishes construct validity for applying deep learning to routine scans for population-level health assessments.
Subject of Research: Opportunistic deep learning assessment of body composition and low back pain
Article Title: Deep learning reveals relationships between muscle and adipose tissue and their associations with age, body mass index, and low back pain: A cross-sectional study in 10,835 subjects
Article References: Marusich, K. R., Gatti, A. A., Blankemeier, L., Boutin, R. D., Smuck, M., Gold, G. E., Delp, S. L., & Chaudhari, A. S. (2026). Deep learning reveals relationships between muscle and adipose tissue and their associations with age, body mass index, and low back pain: A cross-sectional study in 10,835 subjects. PLOS Digital Health, 5(11), e0001788. https://doi.org/10.1371/journal.pdig.0001788
Image Credits: AI Generated
DOI: 10.1371/journal.pdig.0001788
Keywords: computed tomography, deep learning, body composition, low back pain, skeletal muscle, visceral adipose tissue, subcutaneous adipose tissue, cross sectional area, muscle loss, aging, body mass index, electronic health records, opportunistic screening, adipose tissue, matched cohort, musculoskeletal health, abdominal imaging
Cite Scienmag News
Blake Davidson. (November 19, 2026). Deep Learning Study Finds No Bulk Muscle Difference in Back Pain. Scienmag. https://scienmag.com/deep-learning-study-finds-no-bulk-muscle-difference-in-back-pain/
Blake Davidson. "Deep Learning Study Finds No Bulk Muscle Difference in Back Pain." Scienmag, 19 November 2026, https://scienmag.com/deep-learning-study-finds-no-bulk-muscle-difference-in-back-pain/. Accessed 19 November 2026.
Blake Davidson. "Deep Learning Study Finds No Bulk Muscle Difference in Back Pain." Scienmag. November 19, 2026. https://scienmag.com/deep-learning-study-finds-no-bulk-muscle-difference-in-back-pain/








