Topologically Preserved Of Neuron Morphology Analysis

Neuron morphology analysis is one of the key contents of neuroscience research,and it plays an important role in the establishment of the whole brain neuroatlas and the correlation analysis between the whole brain neurons.With the improvement of imaging technology,more and more high-resolution brain image data are collected.However,higher-resolution data will cause excessive data storage overhead and a sharp increase in the complexity of neuron morphological analysis algorithms.The large number of neurons and their complex and diverse shapes pose a huge challenge to neuron mor-phology analysis.At present,a complete Drosophila brain nano-electron microscope scan image has a volume of 40 trillion pixels.Faced with such a large-scale data,it is impossible to rely on manual analysis alone.Therefore,an efficient,convenient and automatic algorithm is needed to complete Morphological analysis of neurons.In order to solve the above difficulties,this thesis perceives the neuron topology and geometry from the two expressions of the skeleton and the point cloud,and analyzes the neuron morphology in the whole brain.The details are as follows:(1)The framework of neuron morphology analysis algorithm based on skeleton expression is designed.The framework first designed a skeleton extraction algorithm that maintains the topology of the neuron in response to the huge amount of original data and the excessive consumption of storage and computing resources.It is expressed through a tree structure to find the root node to the leaf node.The shortest path to complete the skeleton extraction of neurons.Based on the neuron skeleton structure and its distribution in the brain area,hierarchical clustering is used to complete neuron morphological analysis.(2)A morphological analysis algorithm based on point cloud expression is pro-posed.The expression of neuron skeleton can more accurately describe the slender branch structure,but the segmentation results in the Drosophila electron microscope image include a large number of neuronal cells and other biological tissue structures with different shapes in addition to the typical tree-like structure of neurons.It is diffi-cult to extract an effective skeleton expression.Therefore,this thesis extracts the surface point cloud from the dense segmentation results of the Drosophila electron microscope,and completes the morphological analysis of the whole brain biological structure based on the deep neural network to extract the geometric features of the point cloud.Specif-ically,a small number of annotations are used to train the point cloud feature extraction network,and then an unsupervised clustering method is used to automatically cluster the proposed features to complete the morphological classification of the results of the whole brain segmentation of Drosophila.In general,this thesis uses skeleton expression and point cloud expression to an-alyze the morphological analysis of Drosophila whole brain neurons and cell structure from the perspective of neuron topology and geometric morphology.Real-time visual-ization and neuron annotation provide convenient conditions.The morphological anal-ysis algorithm based on point cloud provides a new research idea for neuron tracking of large-scale over-segmentation electron microscopy data.