Research On Automatic Measurement Method Of Spinal Scoliosis Cobb Angle Based On Deep Learning

Spinal scoliosis is a three-dimensional deformity of the spine,characterized by lateral deviation of the vertebrae from the body’s midline in the coronal plane.The etiology of scoliosis is still unclear.Cobb angle is considered the "gold standard" for assessing the severity of spinal scoliosis,and its measurement results are used for treatment planning and disease progression tracking.However,traditional manual methods for measuring Cobb angle in clinical practice suffer from issues such as inter-observer variability,interobservation variability,and low efficiency.Therefore,there is a pressing need for an accurate and efficient automatic Cobb angle measurement tool.To address this need,this study proposes a deep learning-based framework to study the vertebral segmentation,keypoint localization,and engineering implementation involved in automatic Cobb angle measurement.The specific contributions are as follows:(1)Proposed a multi-task deep learning network,MVIE-Net,for feature extraction from spinal X-ray images.First,to obtain original training data samples,this study created segmentation labels and keypoint heatmap labels for the MICCAI 2019 AASCE Challenge public dataset and augmented the data through mirroring,rotation,and other methods.Based on this,MVIE-Net was proposed,which can simultaneously perform vertebral segmentation and keypoint heatmap segmentation tasks.Experimental results demonstrate that the proposed network achieves segmentation accuracy of 85.5% for vertebral segmentation and 41.6% for keypoint heatmap segmentation,surpassing commonly used medical image segmentation networks.(2)Proposed various modeling methods for Cobb angle calculation,including line fitting,corner point method,and point-line mean method.First,based on the segmentation results of MVIE-Net,the upper and lower endplates of the vertebrae were fitted and the key points were parsed and paired.Cobb angle was then calculated using various methods such as midpoint method,line fitting method,corner point method,and point-line mean method,which were proposed and compared.Finally,MVIE-Net and Cobb angle calculation methods were integrated to achieve automatic Cobb angle measurement.Experimental results demonstrate that the proposed automatic Cobb angle measurement algorithm achieves an angle error of ±3.31°,which is currently the best performance on the dataset.(3)Developed a corresponding web application system based on the automatic Cobb angle measurement algorithm.This system not only provides easy-to-use automatic Cobb angle measurement functionality,but also includes Lenke classification for spinal scoliosis.Through this system,users can quickly and accurately diagnose spinal scoliosis.Overall,the experiments show that the deep learning-based automatic Cobb angle measurement algorithm proposed in this study has high accuracy and efficiency,and good generalization performance.The developed automatic Cobb angle measurement system for spinal scoliosis can provide comprehensive and reliable diagnostic services for doctors and patients.