Electron Microscopic Image Segmentation Using Deep Learning

Reconstructing Electron Microscopic(EM) image stacks of mouse brain tissue requires many hours of manual annotation and proofreading, which increases the need for automatic image segmentation methods. The purpose of this work is to conduct automatic segmentation using state-of-the-art deep learning tools available like Tensorflow to train a U-Net architectured neural network to predict the probability of a pixel in the EM image representing a cell boundary membrane. We used a dataset containing 356 EM image stacks with annotations provided. Currently, the tensorflow based network runs on a single Cooley node and work will continue to evaluate the results.