Automatic Spine Curvature Estimation from X-ray Images

Idiopathic scoliosis, a common spinal deformity, poses health risks if not detected and treated early, with the Cobb angle being a crucial diagnostic metric. However, manual measurement of the Cobb angle is prone to inter-operator variability and is time-consuming for clinicians, hindering timely and accurate diagnosis. Existing methods may also not generalize well to unseen X-ray images from different distributions.

Our goal is to create robust automated methods that accurately measure the Cobb angle from X-ray images, adapting to images from distributions not seen during training, and offering clinicians a dependable tool for timely diagnosis.

The project has developed initial automated methods combining vertebra detection and corner regression to estimate Cobb angles. Early results demonstrate reduced inter-operator variability and manual effort, with ongoing work focused on refining robustness across varied datasets and preparing the methods for clinical deployment.

• Presented at MICCAI 2019 CSI Workshop & Challenge
• Published in Lecture Notes in Computer Science (Springer, Feb 2020)