Automating Cobb angle measurement in adolescent idiopathic scoliosis based on leveraging a computer vision algorithm

Introduction The Cobb angle is the gold standard metric for evaluating adolescent idiopathic scoliosis. Currently, measurement is done manually by a physician using scoliosis radiographs, which can be time-consuming and subject to human error. We aim to evaluate the accuracy and efficiency of a computer vision algorithm in measuring Cobb angles for adolescent idiopathic scoliosis compared to manual measurement. Methods A computer vision-based algorithm to automate the process of measuring the Cobb angle was developed using 500 annotated scoliosis radiographs and tested on 54 images. Measurements were compared to those of two orthopaedic surgeons in terms of angle accuracy, detection of multiple curves, and measurement time. The application is run locally with a model built on Tensorflow Keras using Python. Results The algorithm can correctly identify scoliosis, correctly segment multiple curves within the same spine, and calculate a Cobb angle with a mean angle difference of 4.50 + 5.19° compared to the human measurements, within acceptable interobserver reliability. It measured Cobb angles in under 10 seconds in >95% of cases, versus an average of 3 minutes manually, representing a time savings of over 90%. All users found the software intuitive and clinically helpful. Conclusion The algorithm can reliably and efficiently measure multiple Cobb angles in a plain coronal scoliosis radiograph in less than 10 seconds. This can potentially be deployed to improve the accuracy and efficiency of Cobb angle measurements in various healthcare settings and can potentially be used in pre-operative surgical planning and intra-operative alignment checks in the future.