Termination Detection In Three-Dimensional Neuronal Images Using An Adaptive Ray-Shooting Model

Brain is one of the most important organs of organism,but human beings still know little about the fine structure and working mechanism of brain.Researchers can understand the working mechanism of brain by studying the structure and function of brain neurons and their connection networks.A key component in this process is the neuron morphological structure reconstruction from a large data set of volumetric microscopy images.The implementation of neuron morphological structure reconstruction algorithm needs to extract seed points.This article aims at the selection of ideal seed points in 3D neuron morphological structure reconstruction algorithm.Therefore,this paper proposes an algorithm based on adaptive Ray-Shooting model to detect 3D neuron termination,and then the neuron termination is used as seed point,which effectively realizes the reconstruction of the 3D neuron morphological structure.The main research contents are as follows:(1)In the aspect of image preprocessing,this paper calculates the size of the connected region,and the anisotropic filtering is used to achieve the purpose of removing speckle noise from the neuron image and enhancing the blurred foreground of the neuron image.(2)In 3D neuron image stack,a real 3D termination should keep its 'termination'characteristic in the corresponding 2D slice,the 3D neuron termination can be selected from the 2D neuron termination.An adaptive Ray-Shooting model is proposed to detect the neuron termination by analyzing the intensity distribution of the neighborhood around the termination candidate,which effectively realizes the detection of the 2D neuron termination.The radius of adaptive Ray-Shooting model is selected according to the local diameters of the neuron,this paper proposes Multistencil Fast Marching(MSFM)method to calculate the local diameters of the neuron.(3)We find all the 2D termination in the 2D slices as the 3D termination candidate.Then,this paper proposes adjacent slices verification based on a Gaussian function model(the number of slices in the Gaussian function model is selected according to the local diameters of the neuron).By using the function fitting method and calculating the value of mean squared error(MSE),the detection of the 3D neuron termination is realized.The experimental results on Digital Reconstruction of Axonal and Dendritic Morphology(DIADEM)and BigNeuron datasets validate feasibility of our proposed algorithm in this paper.