Neuronal networks reconstruction is rather a challenge in the neuroscience.

  Recent developments in volume electron microscopy (EM) imaging have enabled us to obtain large amounts of brain tissues imaging data.

  Analysis of the tremendously huge electron microscopy (EM) neuronal images based on automated method would be of vital importance.

  In this paper we propose a method that training deep convolutional neural network (DCNN) on labeled data for neuronal boundary detection;

  and then with the membrane detection probability map (MDPM) generated by DCNN, a marker-controlled watershed method is used to segment neurons in the EM images.

  After getting the sequence of 2D EM neuronal images segmented, semi-automated and automated 3D reconstruction methods are employed to

  connect the sections of the corresponding segmentations belonging to each neuron. Finally, we have reconstructed dense neurons in 500 of 1793

  scanning electron microscopy (SEM) images of  drosophila mushroom body with automated method and several neurons with semi-automated method.