High-Precision Automated Reconstruction of Neurons with Flood-filling Networks

• DOI: 10.1101/200675
• Publication Date: 2017-10-10T05:10:31Z
• Authors: Michał Januszewski; Jörgen Kornfeld; Peter H. Li; Art Pope; Tim Blakely; Larry Lindsey; Jeremy Maitin-She...; Mike Tyka; Winfried Denk; Viren Jain
• Keywords: Tracing; Neurite

Abstract Reconstruction of neural circuits from volume electron microscopy data requires the tracing complete cells including all their neurites. Automated approaches have been developed to perform tracing, but without costly human proofreading error rates are too high obtain reliable circuit diagrams. We present a method for automated segmentation that, like majority previous efforts, employs convolutional networks, contains in addition recurrent pathway that allows iterative optimization and extension reconstructed shape individual processes. used this technique, which we call flood-filling trace neurons set obtained by serial block-face male zebra finch brain. Our achieved mean error-free neurite path length 1.1 mm, an order magnitude better than previously published applied same dataset. Only 4 mergers were observed test 97 mm length.