Research On The Reconstruction Of Mitochondrial On Electron Microscope Images

Mitochondria are organelles that exist in most cells.As the main place for the pro-duction of energy in cells,mitochondria and Biological metabolism and heredity have the closely relation.With the development of scanning electron microscope,we have a deeper understanding of the internal structure of the cell.However,the segmenta-tion and reconstruction of mitochondria from EM image is still a challenging problem.Despite the current remarkable algorithms have achieved ideal results,they have poor performance in Neural images.In the past two years,the study of deep learning has been very active,especially in the field of image processing,the neural network has been widely used.In this paper,we present a new strategy which utilizes the Faster-RCNN algorithm for detection and determining the location of the mitochondria in the image.And then use the multi-layered information fusion for the connection across adjacent sections.the feature matching algorithm and manual correction are adopted to improve accuracy in the process.In this paper,the cerebral neuronal serial section images are obtained from platfor-m of microcosmic reconstruction and analysis group in Institute of Automation,Chi-nese Academy of Science.In the process of data acquisitiorn,1mm3 murine brain tissue produced about 2000TB of nerve size images on the nanometer scale.During the entire reconstruction process,the microscope images in multiple different views need to be spliced into a three-dimensional image containing complete neurons.Then the mito-chondria are extracted and transformed into a mathematical model that is easy to ana-lyze,and the resulting data showed explosive growth.we use the Hadoop+hbase solu-tion to solve the problem of large scale data access and operation processing in the mi-tochondrial reconstruction process.At the same time,we use three-dimensional recon-figuration software ImageJ for sample labeling,manual correction,three-dimensional display.In order to improve the efficiency of reconstruction,we improve the function of image access in plugin Trakem of ImageJ,and seamlessly interface with the software platform to enable it to access large-scale images in the storage system in real time.