Critical Point Detection For Tree-like Structure Images Using A Ring-like Ray-shooting Model

The morphology reconstruction of tree-like structures plays a significant role in many biomedical and geography studies.To study the morphology of tree-like structures in biomedical and road images,digital reconstruction of tree-like structures is required.The correct location and identific ation of critical points,including terminations,branch points and cross points,for tree-like structures in images is a crucial task in many tree-like structure reconstruction processes.The critical points can not only provide topological information to improve the reconstruction results,but also are very important for the pathological analysis of cell morphological evolution and demarcation-related diseases and the construction of brain atlas.Since few methods have been proposed to detect all three t ypes of critical points simultaneously,we propose a 2D critical point detection algorithm based on a ring-like ray-shooting model to detect all types of critical points in tree-like structure images simultaneously.In addition,since most of the biomedica l images are 3D images,we extend the proposed 2D detection algorithm to 3D detection algorithm.The main research contents are as follows:First,we proposed a method of 2D critical point detection in tree-like structure images based on a ring-like ray-shooting model.Given an input image,which is enhanced and segmented to obtain the segmented image,an existing skeleton extraction approach is employed to provide the candidate critical points in the segmented image.Then,a ring-like ray-shooting model is used to extract the pixel intensity distribution feature along the potential branches around the candidate critical points.The ring-like ray-shooting model is a pixel extraction model that can accommodate the size diversity of different candidate critica l regions and the density of the neighborhood branches around the candidate critical points.The inner and outer radii of the ring-like ray-shooting model are respectively measured by the radius estimation algorithms based on a sphere-growing approach and a modified ray-burst model.The DBSCAN algorithm is subsequently adopted to complete the detection of 2D critical points for tree-like structures by analyzing the number of clusters of pixel intensity distribution extracted from the ring-like ray-shooting model.Second,we extend the proposed 2D detection algorithm to 3D detection algorithm.The 3D images are firstly sliced into a group of 3D sub-images in multiple directions,and 3D sub-images are projected with maximum intensity projection(MIP),which can be called multiple maximum intensity projection.The 2D branch points are subsequently detected in the MIP images using the proposed 2D detection method based on a ring-like ray-shooting model.Then,by using a reverse mapping approach,the 2D branch points with the same 2D coordinates and pixel intensity are mapped to 3D sub-images,and the 3D branch points of each sub-image are obtained.Finally,the real 3D branch points in original images are de tected by fusing the 3D branch points in 3D sub-images.This strategy can not only solve the problem of resolution inconsistency,but also improve the detection efficiency of the algorithm.The experimental results on 430 2D and 59 3D images confirmed tha t the proposed method can achieve very accurate 2D critic al points and 3D branch points detection results,and has a competitive performance compared with the state-of-the-art detection method.In addition,the proposed algorithm is dozens of times faster than the existing deep-learning based model under the same computer configuration.The proposed method has excellent performance in both detection precision and computation efficiency for junction detection even in large-scale biomedical images.