Algorithm And Proof Tool Of Neuronal Electron Microscopy Image Segmentation Based On Super-pixels

Neuronal microstructural reconstruction is the basis on exploring the principal of cerebral nervous tissue.To reconstruct cerebral neuronal microstructure,we first need to obtain the samples of creatural cerebral nervous tissue.Then,electron microscopy(EM)is used to gathering serial section images of these samples.Finally,we apply neuron segmentation methods and tools which based on machine learning and computer vision algorithms to processing and analyzing the images of nervous tissue acquired by electron microscopy.After that,we can get the informatic representation of the neuronal microstructure.Neuron segmentation is the basic topic of neuronal electron microscopy image processing,which is also a big challenge in neuronal microstructural reconstruction.In most current researches,semi-automatic method which contains automatic segmenting algorithms and manual labeling software are widely used to solving the problem of neuron segmentation.In this paper,the cerebral neuronal serial section images are obtained from platform of microcosmic reconstruction and analysis group in Institute of Automation,Chinese Academy of Science.We proposed an algorithm used for automatic neuron segmentation based on super-pixel,and developed software modules for manual verification.In automatic algorithm.Firstly,we utilize deep convolutional neural network to detect membrane in neuronal EM images.Then,super-pixels are producing by watershed algorithm.Last but not least,we use hierarchical structure to merge these super-pixels to get final results.The function of manual verification software module is modifying the automatic segmentation results.Image data will be stored in text files to get higher performance.Modifying module is developed based on the open source software Trak EM2.