Research On Computer Aided Measurement Algorithm For Cobb Angle Of Scoliosis Image

With the development of computer vision and image processing technology,new computer-aided diagnostic methods have been provided for medical image such as X-ray,MRI,CT and PET.It is significant to explore a computer-aided measurement of Cobb angle that relies less on prior knowledge and personal operations and can obtain more stable measurement results.Our paper mainly studies X-ray,CT and MRI images of patients with scoliosis provided by a hospital in Anhui Province.The following aspects are studied:(1)An automatic Cobb angle measurement algorithm based on traditional image processing algorithms for MRI and CT scoliosis images is proposed.Firstly,the enhanced watershed segmentation algorithm is proposed to segment the MRI or CT spine image and extract the spine center points.Then,the sixth-order polynomial is used to fitting spine curve by center points.Finally,the Cobb angle is automatically calculated by calculating the angle between the tangent of the point in the spine curve where the second-order derivative is zero.The experimental results show that the algorithm can achieve 94.2%of the spine segmentation accuracy and 40°Cobb angle measurement error.The Cobb angle measurement can be calculated automatically based on the proposed algorithm without the manual definition of upper and lower end-vertebrae in the current Cobb angle computer aided measurement algorithm.(2)An automatic traditional machine-learning based Cobb angle measurement algorithm for scoliosis is proposed for X-ray images.Firstly,the aggregated channel features with higher feature performance and effective description of the target are introduced.The multi-scale aggregated channel features are extracted from the spine X-ray image and the spine region detection is performed by using the Adaboost classifier to train cascade classifier.The segmentation algorithm based on neighborhood information and intensity value is used to segment the spine contour of the spine region.Finally,the spine curve is fitted and the Cobb angle is automatically calculated.The experimental results show that the algorithm can achieve 98.5%detection accuracy,a segmentation accuracy of 80.33%and 4.99° Cobb angle measurement error.The image category of the Cobb angle automatic measurement of the scoliosis image is expanded,which is suitable for X-ray scoliosis images with wider clinical application.(3)A spine detection and segmentation algorithm DU-Net based on deep convolutional neural network for X-ray images is proposed.Firstly,the above-mentioned spine detection algorithm is introduced to construct a spine detection model.Then the semantic segmentation network U-Net is used as the spine contour segmentation framework to construct the spine segmentation model,and combined with the spine detection model to form the DU-Net detection and segmentation network.The experimental results show that the algorithm can achieve an average Dice coefficient of 90.28%,a segmentation accuracy of 86.3%and an IOU of 82.29%.The accuracy of the spine contour segmentation of the X-ray image is improved,and it is suitable for the spine contour segmentation module in the Cobb angle automatic measurement algorithm.