Identifying scoliosis in a population-based adult cohort: automation of a validated method based on total body dual energy X-ray absorptiometry scans

Scoliosis is a three-dimensional spinal deformity that, if left untreated, can progress to painful disability and require invasive surgical correction. In this work, we propose an algorithm to automate the established manual DXA Scoliosis Method (DSM) for measuring scoliosis in adults, that simultaneously quantifies curve angle, direction and apex location. The automation pipeline comprises two stages: (i) segmentation of the spine and computation of its spine curve, and (ii) geometric analysis of that curve to identify the apex and derive the maximum modified-Ferguson angle as a continuous measure of severity. The automated method is validated for accuracy and repeatability. The validation is by a direct comparison with manual DSM measurements on 1,929 UK Biobank scans. The repeatability is from a test–retest analysis on a second DXA scanning session acquired 12 months after the primary scanning session for 2,728 participants. Automated angles show excellent concordance with manual readings (Pearson r = 0.9; median bias = -1.8 $$^\circ$$ , 95% limits = $$\pm 5^\circ$$ ). Classification performance for key curve descriptors is high: location (specificity = 0.9), direction (0.8), and single- vs double-curve type (0.7). Validation between two sessions for the same patients reveals good agreement with a mean difference between sessions of 0.2 $$^\circ$$ . Given that DXA scanning is fast, inexpensive, and easily collected in large cohorts (e.g. UK Biobank and Avon Longitudinal Study of Parents and Children), the automated DSM enables population-scale spinal curvature phenotyping. This advancement will support robust scoliosis epidemiology, genetics, and natural history, potentially informing evidence-based screening strategies.