To create a machine learning (ML) model for auto mated Cobb angle calculation on spine radiographs in paediatric patients with suspected idiopathic or congenital scoliosis. 1
MethodologyA retrospective cohort study was performed over a 10-year period in 130 paediatric patients with suspected idiopathic (group 1) or congenital (group 2) scoliosis. Cobb angles were measured both manually by radiologists and by a ML segmentation-based approach using an Augmented U-Net model. Symmetric Mean Absolute Percentage Error (SMAPE) was calculated, whereby a lower SMAPE corresponded to a better performance of the ML model.
ResultsThe Augmented U-Net ML model achieved an overall SMAPE of 11.82% for the combined idiopathic and congenital scoliosis population. In the idiopathic scoliosis group, a SMAPE of 13.02% was achieved, while the congenital scoliosis group achieved a SMAPE of 11.90%.