Multicut brings automated neurite segmentation closer to human performance

Constantin Pape

Abstract: The connectomics community is acquiring volumetric electron microscopy (EM) images of the brain at an unprecedented rate with the aim of mapping out and understanding in detail the physical correlates of information processing in animals. Reliable automatic segmentation is urgently needed for upcoming whole-brain data sets (>100 terabytes (TB) per volume). Manual analysis, despite impressive progress in collaborative annotation1, will not scale to this massive task. We present an algorithm and software package to segment such data sets with low error rates.