Research on the detection method of full spine X-ray based on deep learning
Objective Accurately locating the boundaries of the entire spine is a prerequisite for the correct evaluation and treatment of most spinal diseases.At present,deep learning technology has been gradually introduced into the detection task of the spine and has achieved good results.However,there are few studies on automatic detection of the entire spine.Based on the anteroposterior and lateral spinal X-rays,this paper realizes the detection of the vertebral bodies of the entire spine,and then realizes the automatic classification of lumbar spondylolisthesis of the spine based on the detection results.Methods In this paper,a target detection framework,is used to detect the vertebral body,arch root,upper edge of the sacrum S1,and the center of the femoral head.Then,clinical parameters are calculated based on the test results to complete the classification prediction of the spine.Results The data set used is from the 8331 X-rays of the partner Peking University Third Hospital,and each patient includes an anteroposterior view and a lateral view.For vertebral body detection,this method achieves the recall rate of 99.46%and the precision rate of 99.73%in the anteroposterior view,and the recall rate of 98.83%and the precision rate of 99.38%in the lateral view.The results of the detection of the upper edge of S1,the center of the femoral head,and the arch root met the requirements of clinical experts.The average accuracy of the classification of lumbar spondylolisthesis was 93.5%.Conclusion This study provides a deep learning-based solution to the problem of lumbar spondylolisthesis.This method can also be extended to other spinal detection problems such as cervical spondylolisthesis,scoliosis,and vertebral anomalies.
万方数据
维普
