A Multiview-based Automatic Algorithm For 3D Fracture Line Extraction

With the improvement of digital three-dimensional medical technology,orthopedists hope to extract the fracture lines of the patient’s fracture for fracture treatment,analysis and teaching.However,no effective automatic extraction method exists,and fracture lines still require manual work of doctor.Obviously,there are many difficulties in the automatic extraction of fracture lines.For example,fracture fragments reconstructed by CT(Computed Tomography)have high noise,complex shapes,holes and other defects.To deal with the challenge of automatic extraction of fracture lines,this thesis designs an automatic pipelines for fracture line extraction.Its purpose is to automatically extract the surface fracture lines of multiple reassembled fracture fragments and map them to the surface of the bone template.Specifically,the main work done in this thesis includes:(1)An outer surface generation algorithm for complex 3D meshes is proposed.The algorithm firstly culls redundant vertices and triangles based on occlusion detection,and then reconstructs the outer surface for complex meshes through mesh repairing.Experiments show that the outer surface generation algorithm is fast and robust,which can provide technical support for the generation of bone templates and provide assistance to other algorithms proposed in this thesis.(2)An automatic extraction algorithm of complex 3D fracture lines based on multiview is proposed.Based on multiview rendering and point cloud centering,the algorithm can quickly and accurately extract surface fracture lines for multiple reassembled fracture fragments.First,combined with spherical uniform sampling and clustering,we design an adaptive multiview selection strategy to select an appropriate number and location of rendering views for each set of input fracture fragments.Next,the algorithm samples the surface of the fracture fragments based on off-screen rendering,and selects the samples around the fracture lines through a new framebuffer-based crack recognition method.Finally,we propose a new point cloud centering method based on smoothing and merging to extract fracture lines on the surface of fracture fragments.Experiments show that the algorithm can quickly,accurately and robustly extract fracture lines on the surface of complex fracture fragments,and can generalize to the extraction of crack lines on any classified mesh surface.(3)An automatic mapping algorithm for complex 3D fracture lines is proposed.First,we registered the outer surface of the fracture fragments with the bone template.Then,we propose a normal mapping algorithm to attach the fracture lines to the adjacent mesh surface to assist the attachment of the fracture lines to the target mesh surface.Finally,we design an step-by-step deformation algorithm based on the polyharmonic deformation to make the deformed outer surface as close as possible to the shape of the bone template.Experiments show that the mapping algorithm is efficient and relatively reliable,and can effectively help doctors to achieve similar fracture analysis needs such as making fracture maps.A variety of experiments have been designed in this thesis,and the experiments have proved that our algorithm can be applied to multiple groups of complex fracture fragments,and has fast speed and strong practicability,which can effectively help doctors save a lot of time.Additionally,the algorithm in this thesis can not only achieve fracture line extraction on the surface of complex 3D fracture fragments,but also extend it to the subject of crack detection and crack line extraction between classified 3D meshes.