Automated Cobb Angle Measurement at Scale: Identifying Undiagnosed Scoliosis in the UK Biobank

AbstractBackground contextAdult degenerative scoliosis develops after skeletal maturity in a previously normal spine, often as a result of age-related spinal degeneration, with its prevalence increasing due to longer life expectancy. Structural changes such as disc degeneration, osteoporosis, and measures of muscle quality can contribute to spinal instability. The Cobb angle is the standard method for quantifying spinal curvature severity, calculated as the angle between the most tilted vertebrae.PurposeTo derive automated Cobb angle measurements from chemical-shift-based water-fat separation (Dixon) magnetic resonance images (MRI) and investigate associations with anthropometric, socioeconomic variables, health conditions including back pain, and other image-derived phenotypes related to body composition.DesignCross-sectional study.Patient sampleParticipants from the UK Biobank study having abdominal MRI scans from the first imaging visit (n = 33,889).Outcome measuresThe maximal Cobb angle was automatically derived from vertebral bone marrow segmentation via a centroid-based spline-fitting method.MethodsWe measured the Cobb angle from MRI scans using a deep learning algorithm for vertebral bone marrow compartment labelling, followed by spline fitting over these labels. We compared scoliosis prevalence based on both self-reported as well as clinically diagnosed scoliosis and MRI-derived Cobb angle defined scoliosis. Additionally, we explored sex-specific relationships between Cobb angle and anthropometric variables, body composition measurements, and health conditions using correlation and linear regression modelsResultsFrom a cohort of 33,889 UK Biobank participants, scoliosis was formally diagnosed or self-reported in a total of 170 subjects (0.5%), whereas 9,497 participants (28%) were identified on abdominal MRI via the algorithm as having some form of scoliosis based on a Cobb angle >10°. Among participants without a formal diagnosis of scoliosis, but identified by the algorithm, the majority (95.4%) were categorised as having mild scoliosis (< 25°). Compared to males, female participants had a higher mean Cobb angle (9.02 ± 5.59° vs 8.44 ± 4.79°, respectively), were more likely to have scoliosis, and exhibited higher Cobb angles across all age groups (p < 0.000549, Bonferroni-corrected). Linear regression analysis revealed significant associations between Cobb angle and age (β = 0.12 [0.11, 0.13] for female, β = 0.09 [0.08, 0.10] for male), paraspinal muscle fat infiltration (β = 0.08 [0.07, 0.08] for female, β = 0.05 [0.04, 0.06] for male), chronic back pain (β = 0.07 [0.04, 0.09] for female, β = 0.05 [0.02, 0.07] for male), and visceral adipose tissue (β = -0.06 [-0.08, -0.05] for female, β = -[-0.07, -0.05] for male). Additionally, iliopsoas muscle volume was significantly negatively associated with Cobb angle in male participants only (β = -0.02 [-0.03, -0.01], p < 0.000549). However, the sex-stratified models showed low R² values, indicating that other unmeasured factors substantially contribute to the observed variation.ConclusionsThese findings suggest that a significant portion of the UK population has scoliosis that remains undiagnosed clinically, the majority of whom have a mild form of scoliosis. Our results are consistent with previous studies showing a higher prevalence of scoliosis in females as well as highlighting the association between spinal curvature, paraspinal muscle quality, and chronic back pain regardless of gender, osteoporosis or age.