Cross-sectional area and fat infiltration of the lumbar spine muscles in patients with back disorders: a deep...

Validated deep learning models represent a valuable option to perform large-scale research studies aiming to evaluate muscle quality and quantity of paravertebral lumbar muscles at the population level. This study aimed to assess lumbar spine muscle cross-sectional area (CSA) and fat infiltration (FI) in a large cohort of subjects with back disorders through a validated deep learning model.

T2 axial MRI images of 4434 patients (n = 2609 females, n = 1825 males; mean age: 56.7 ± 16.8) with back disorders, such as fracture, spine surgery or herniation, were retrospectively collected from a clinical database and automatically segmented. CSA, expressed as the ratio between total muscle area (TMA) and the vertebral body area (VBA), and FI, in percentages, of psoas major, quadratus lumborum, erector spinae, and multifidus were analyzed as primary outcomes.

Male subjects had significantly higher CSA (6.8 ± 1.7 vs. 5.9 ± 1.5 TMA/VBA; p < 0.001) and lower FI (21.9 ± 8.3% vs. 15.0 ± 7.3%; p < 0.001) than females. Multifidus had more FI (27.2 ± 10.6%; p < 0.001) than erector spinae (22.2 ± 9.7%), quadratus lumborum (17.5 ± 7.0%) and psoas (13.7 ± 5.8%) whereas CSA was higher in erector spinae than other lumbar muscles. A high positive correlation between age and total FI was detected (r_s = 0.73; p < 0.001) whereas a negligible negative correlation between total CSA and age was observed (r_s =  - 0.24; p < 0.001). Subjects with fractures had lower CSA and higher FI compared to those with herniations, surgery and with no clear pathological conditions.

CSA and FI values of paravertebral muscles vary a lot in accordance with subjects' sex, age and clinical conditions. Given also the large inter-muscle differences in CSA and FI, the choice of muscles needs to be considered with attention by spine surgeons or physiotherapists when investigating changes in lumbar muscle morphology in clinical practice.