A Study Of Neurite Tracing Method In Multi-neuron Data

The analysis of the morphological structure of neurons is one of the important methods in the study of brain.The advanced microscopic imaging technology has provided largescale neuron data.The digital reconstruction of neuronal morphology is a key method for neuromorphological analysis.Neurite tracing is the main step of neuronal morphological reconstruction.Although many automated Neurites algorithms have been proposed,most of them focus on single neuron data tracing.,Fast and reliable algorithm that can replace the time-consuming and labor-intensive manual tracing process in multi-neuron data remains an unsolved problem.Among many factors,the problem of weak signal extraction and the determination of the topological relationship of neurons are considered to be the major bottleneck in the tracing of multi-neuron data.We study the image enhancement methods of neurite on the extraction problem of weak signals.Firstly,we implemented the multi-scale operator algorithm as an image enhancement method.Then we developed a new image enhancement method that based on the Full Convolutional Neural Network Transfer Learning,which provided a new way to improve signal integrity in neurite threshold segmentation.We developed a Classification and Re-tracing method to solve the problem of neuron topological errors.The method mainly includes two parts: the classification module and the re-tracing framework.In the classification module part,the initial tracing results were classified by the support vector machine algorithm to correct the topological errors in the initial tracing dispersedly.The module is composed of region extraction,fragment extraction and fragment classification.It achieved a classification result that equivalent to manual work when the signal intensity of neural is low.In the re-tracing framework,the initial neurite tracing method was implemented based on the voxel coding algorithm.And the secondary tracing was implemented with partial data replacement operation.The results of two parts were integrated to form a new multi-neuron data neurite auto-tracing method.