The Research On Three-Dimensional Reconstruction Of Scoliosis Based On 2.5-Dimensional Extended Field-of-View Ultrasound

The examination and diagnosis of scoliosis extremely rely on medical imaging technology.The X-ray is the most commonly used examination method in clinics at present.Doctors calculate the Cobb angle to assess the severity of scoliosis.However,due to the radiation of X-rays,it is harmful to the health of human body and is unsuited for multiple inspections in a short time.In addition,manual measuring Cobb angle will cause a burden on doctors,and latent errors may exist.Therefore,researching on a computer-assisted diagnosis system for scoliosis is particularly necessary.Ultrasound imaging is widely used in clinical examination and diagnostics,because of its advantages of no radiation,low cost and high portability.The emergence of ultrasound extended filed-of-view(US EFOV)imaging technology has solved the drawback that ultrasound imaging is limited to a small field of view,and greatly expands the application of ultrasound imaging.However,traditional 2D EFOV images lack stereoscopic spatial information,and 3D EFOV imaging also has drawbacks,such as time-costing,low resolution and higher hardware requirements.In order to overcome the shortcomings of traditional EFOV technology,2.5-dimensional(2.5D)EFOV based on the positional information has emerged.In this paper,we applied the 2.5D EFOV technique to the clinical examination of scoliosis.A bilateral scanning method for scoliosis patients and an approximate Cobb angle measurement method based on coplanar circles are proposed.At the same time,in order to provide doctors with a more intuitive 3D structure of scoliosis,we designed a 3D reconstruction system of scoliosis.The 3D reconstruction system recognizes and locates the transverse processes on the 2.5D EFOV images,and uses the 3D position of transverse processes to reproduce the spine model through spatial geometric transformation.Besides,it provides the Cobb angle measurement between any vertebrae.We proposed two methods for segmentation and localization of bone surface of transverse processes on 2.5D US EFOV images.The first method completes the segmentation and localization of bone surface on original frames.An edge-based template matching and full convolution neural network are combined for the bone surface segmentation.At the last step,we use the hierarchical clustering to de-duplicate the redundant locations from segmentation and produce the final location of the transverse processes.The second method combined the bilateral filtering based on 3D position information with fast FCM clustering algorithm and morphological processing to complete the segmentation of 2.5D EFOV images.After the segmentation,we find the peak of borderline of bone surface and its shadow area as the position of the transverse processes using the second-order differential based on distance threshold.In this paper,clinical experiments were conducted to validate the feasibility and accuracy of the 3D spine structure reconstruction method based on 2.5D EFOV technique.The accuracies of two methods for bone surface detecting were compared and the final reconstructed spine structures are compared with the X-ray films.In addition,the accuracy of Cobb angle measurement is validated.The experiment results show that the 3D reconstruction of scoliosis method based on 2.5D EFOV US imaging has high accuracy,visibility,interaction and is potential for clinical utilization.