Neuron Branch Point Detection Method Based On Virtual Cell Imaging And Deep Learning

This paper mainly deals with the problem of branch point detection in auto-matic morphological reconstruction of neuron images.The automatic reconstruc-tion of neuronal morphology is an important means of studying the principles of nervous system and even exploring the mysteries of brain.The existing most com-monly used and effective reconstruction algorithms,local tracing algorithm,rely on the detection of branch points in neurons,which limits the accuracy and speed of algorithm.Based on the virtual cell imaging and deep learning technology,this paper proposed a method for detecting the branch points of the neuron image,which effectively solved the problem.The main work of this paper is(1)The simulation and rendering algorithm of the neuron model is designed and implemented.Based on the cell optical imag-ing model under microscope,we introduce point spread function,noise model and down sampling method to simulate the photoelectric effect and digital to analog conversion process of neuron images.The experimental results show the effective-ness of the algorithm.(2)An algorithm for detecting and locating the branch points of real neurons based on 3D convolution neural network is designed.Ex-perimental results show that the new algorithm can not only accurately detect branch points and locations in real neural images,but also effectively improve re-construction efficiency when integrating them into existing neuron morphological reconstruction algorithms.(3)A neuron morphology refine network nsrGAN is designed,which can optimize the neuron image while minimizing the gap between input image and output image,so as to ensure the optimized image label remains unchanged.