The Research On Three-dimensional Reconstruction Of Scoliosis Based On Object Detection And Clustering

Early detection is significant for scoliosis.Medical imaging technology is used to evaluate the severity and development of scoliosis,which will be favorable for doctors to make decisions on the treatment of patients.Clinically,X-ray is commonly used as imageology examination for scoliosis patients and doctors calculate Cobb angle to evaluate the severity of scoliosis.However,the radiation of X-ray is harmful to people's health and so it is not suitable for the scene of multiple inspections,and the manual measurment way of Cobb angle easily leads to measurement errors.Ultrasound imaging is a radiation-free,low-cost,portable and popular medical imaging technology,which has been widely used in medical examination,diagnosis and treatment in clinical practice.Due to the limited field of view of traditional ultrasound imaging,it often expands its application range by means of other extend field of view technologies such as two dimension,three dimension and 2.5 dimension ultrasound imaging.Whereas,two dimension ultrasound imaging can not obtain three dimension information accurately only by images in the plane.Three dimension ultrasound imaging faces the problem of time-consuming reconstruction,low resolution and high demand for hardware,while 2.5 dimension ultrasound imaging requires high skills for scanning way and it's hard to obtain a specific object in the ultrasound image during the scanning process.Thus it is necessary to research and develop a computer-aided assessment system for scoliosis which is able to automatically locate the position of vertebra landmarks and is able to visualize the spine from multiple perspectives in three dimension space with high interactivity.In this paper,we proposed a scoliosis vertebra landmarks localization system based on object detection and clustering and as the same time we have designed and developed a three-dimension visualization and measurement system for scoliosis.First of all,we utilize object detection algorithm based on deep learning to locate the transverse process and spinous process in ultrasound image,and then the position data was processed by clustering algorithm to get the clustering centers as the final position information,which achieved the goal of locating the vertebra landmarks from coarse to fine.Next,the system will calculate position and posture parameters for each detected vertebra using computer graphics geometry theory and with the help of VTK,vertebra models will be adjusted and connected to be combined into a three-dimensional spinemodel according to the position and posture parameters.In this paper,the corresponding phantom experiment and clinical experiment are designed and conducted.The experiment results have shown that the measurement error of the proposed system is small and it can provide doctors with intuitive and multiple perspectives to observe with high interactivity and it has potential high clinic utility value.