Prediction of the 3D Spinal Alignment from External Shape of the Back in AIS Patients Using Regression Model

One of the most challenging tasks in non-invasive diagnosis of Adolescent Idiopathic Scoliosis (AIS) is to find the correlation between internal spinal deformity and external shape of the back and to create a model to predict internal curve from 3D scans of the back surfaces only. Consequently, X-ray imaging and analysis could be potentially reduced which lead to decreased cumulative dose to the patient and the risk for malignancy development later in life. We proposed a regression model that considers correlation between 3D coordinates of the ground-truth internal spinal curve and 3D coordinates of external asymmetry curve to predict internal spinal alignment close to the real one. With a limited number of samples used to create it, this model is able to provide an acceptable approximation of the spinal internal alignment particularly in sagittal plane, with a mean 3D displacement between predicted and real curve of around 10-12 mm. Quality of prediction could be significantly improved by simultaneously acquiring 3D scans of the back surface and X-ray scans of the patient, and adding to the model external features like scapulae or rib cage which are future steps towards improvement and clinical validation.