Research On Automatic Repair Of 3D Neuron Reconstruction Based On Topological Feature Points

The digital reconstruction of neurons is an important part of neuroscience research.It is very important to understand the connection relationship of nervous system and the construction of the brain atlas.At present,neuron reconstruction mainly relies on manual or semi-automatic reconstruction,which is one of the most time-consuming and labor-intensive parts in the process of neuron data analysis.It greatly hinders the in-depth mining,analysis and processing of neuron data.Due to the existence of gaps in the neuron image,the tracing results generated by automatic reconstruction algorithms may be incomplete,resulting in missing reconstruction branches,so it needs to be repaired during post-processing.Therefore,this paper propose a post-processing algorithm for automatically detecting and repairing the untraced branches of the initial reconstruction,which is based on a multiscale upgraded ray-shooting model(MUR)and a MOST-based(Micro-optical sectioning tomography)repairer.The main research contents are as follows:Firstly,MUR is proposed to detect the missing branches around the topological feature points which contain the junction points detected from the neuron image and the tip nodes extracted from the initial reconstruction.The branch points are detected by the upgraded ray-shooting model by rejecting the surrounding irrelevant pixel information around the candidate points.MUR is proposed to calculate the branch direction vector around the topological feature points to detect the untraced branch and provide seed points for the subsequent repairing process.Specifically,if a branch direction vector points to the untraced region,it is regarded as a untraced branch direction vector,and the position pointed by the untraced branch direction vector is regarded as the seed point of the subsequent repair process.Secondly,a MOST-based repairer is proposed to repair the detected untraced branch.Compared with the original MOST tracing algorithm,on the one hand,the detected branch direction vector is set as the initial reconstruction direction in the MOST-based repairer.This method prevents redundant reconstruction by rejecting the nodes that return to the traced area.On the other hand,a crossing gap strategy is used to complete the reconstruction in the discontinuous area,which extracts the foreground voxels along the trajectory to march forward.When a gap is encountered,if foreground voxels can be extracted in the cone region along the traced trajectory.The centroid of the new voxel cluster will be calculated as a new node.Finally,we find that the proposed method can effectively repair the untraced branches by evaluating five different tracing methods on Big Neuron dataset and SNIWMB dataset.The results of quantitative analysis experiment and visual comparison both show that all the performance metrics are improved after repairing,and the proposed method is helpful to generate more reliable reconstruction results.Besides,the proposed method has been implemented as a plug-in in Vaa3 D platform which is used for biomedical image analysis.Users can effectively repair the false negative neuron reconstruction and visualize the results.