Computation and Prediction Of Cobb's Angle Using Machine Learning Models

Scoliosis is a birth defect that causes the spine to twist out of form. Scoliosis measurement necessitates the labelling and identification of vertebrae in the spine. The most cost-effective and generally available way of imaging the spine is X-rays. Vertebrae segmentation in spine radio-graphs is crucial for image-guided spinal examination, sickness diagnosis, and therapy planning. Traditional assessments rely on time-consuming man-ual measurements that are subject to inter-observer variability. There is no fully automated method for accurately detecting and segmenting the relevant vertebrae in the literature. We provide an end-to-end segmentation model that employs a properly tuned U-Net model with progressive side outputs to enable completely automated and reliable segmentation of the vertebrae associated with scoliosis measurement, followed by our custom code to calculate the Cobb Angle, and finally running a progression model to predict the curvature. Our experimental results from a set of anterior-posterior spine X-Ray images indicate that our model promises to be an effective tool in the identification and prediction of Cobb Angle, eventually helping doctors in the reliable estimation of scoliosis. Moreover, this gives patients a reliable timeline for spinal fusion surgery which is the only reliable cure at present.