A Multidimensional Scoliosis Assessment Algorithm Based on Back Images

Abstract:
Scoliosis refers to the deviation of the spine from the centerline of the torso on one or both sides, forming a "C" or "S" shaped curvature. The rehabilitation period for scoliosis is long and requires regular follow-up. If the patient is frequently exposed to X-rays, it can cause irreversible damage to the blood and reproductive systems. The main research work of this article is as follows:1. Research on monocular ranging algorithm: In response to the problem that the scoliosis assessment module need to measure the distance between the camera and the human body, a technology using a common camera combined with an object detection algorithm is proposed, which can achieve monocular ranging without the need for a depth camera. The relative ranging error is less than 10%, providing quality control function for input images for subsequent scoliosis assessment task. 2. Research on Multidimensional Assessment Algorithms for Scoliosis: In response to the problem of previous studies only focusing on AI assessment of the coronal plane of the spine, this study designed and implemented: (1) an angle of trunk rotation measurement algorithm for assessment the degree of spinal deviation in the sagittal plane, with a relative measurement error of 6.2%; (2) The spinal curvature type discrimination algorithm used to determine whether the spine is in a "C", "S", or multi curvature shape has a discrimination accuracy of 0.887; (3) The spinal coronal scoliosis grading algorithm, which combines doctor's assessment experience and is used to evaluate the severity of the spine on the coronal plane, has a grading accuracy of 0.793, higher than existing methods. This algorithm solves the problem of patients having difficulty performing spinal dynamic assessment at home.