Spinal curvature (scoliosis) affects 3-6% of the population. Scoliosis generally appears in children while they are growing. The curves can get worse both while growing and more slowly with aging. New scoliosis also occurs in adults (called degenerative scoliosis). Large curves are associated with significant problems and may require surgery. Little is known about the causes of curve onset, progression and prognosis. What is needed is a clinical prediction tool to decide which people with scoliosis are most likely to progress and need ongoing monitoring, and who can be reassured and discharged. However, development of a prediction tool needs large scale population-based research to identify predictors. This has been limited so far, because the traditional method of identifying scoliosis is to carry out spinal X-rays which impart a high dose of radiation, and are not ethical for screening the whole population.

To address this, we have developed and validated a method of measuring scoliosis from standard total body dual energy X-ray absorptiometry (DXA) scans, a low-radiation technique using data from the Avon Longitudinal Study of Parents and Children (ALSPAC), a population-based birth cohort. We have then applied our DXA Scoliosis Method (DSM) to 7000 DXA scans from the ALSPAC aged 9 research clinic and 5000 DXA scans from the aged 15 research clinic, and have identified novel associations between early life physical activity and body composition with onset of scoliosis by aged 15. However, the prevalence of scoliosis at aged 15 is relatively low (7.9% in ALSPAC) and to carry out appropriately powered epidemiological studies we need to combine data from multiple research cohorts. We have identified additional research cohorts that have total body DXA scans already performed (approximately 84,500 individuals). We plan to use these cohorts to generate a prediction tool for scoliosis onset and progression. However, our manual method for identifying and measuring scoliosis from DXA scans is labour intensive, and further research is not feasible until we automate this method.

We therefore are proposing development and validation of a fully automatic system to identify and measure spinal curvature from total body DXA scans based on our manual method. The engineering department at the University of Oxford has already developed a similar automated system for three-dimension images of the spine based on magnetic resonance imaging (MRI) scans. We plan to modify this system to allow automatic collection of data on spinal scoliosis from total body DXA scans for future research purposes. This modification process will have two stages: (1) development of a new software algorithm to extract the required features and classify spinal images based on a subset of anonymised DXA images from ALSPAC; and (2) validation of the software on a further dataset of anonymised images from ALSPAC.