Home // BRAININFO 2016, The First International Conference on Neuroscience and Cognitive Brain Information // View article

Incorporate Deep-Transfer-Learning into Automatic 3D Neuron Tracing

Authors:
Zhihao Zheng
Pengyu Hong

Keywords: Neuron Tracing; Convolutional Neuron Network; Neuron Tracing; Deep Learning, Transfer Learning

Abstract:
Automated neuron tracing from microscopic images enables high-throughput quantitative analysis of neuronal morphology to elucidate functions of neural circuits. We have developed a transfer-learning approach that trains a deep convolutional neural network to trace neurons in 3D image stacks. Our neural network model consists of two major components. One is responsible for detecting foreground, the other takes the output of the first components and detect the central lines of neurites. They are trained sequentially, which is more efficient than training a whole deep neural network from scratch. The most spectacular aspect of our approach is that our training data is generated by synthesizing 2D simple lines in noisy backgrounds instead of consisting of manually labeled real neuron images which are labor intensive and time consuming to collect. Our method first processes each slices of 3D image, and then integrate them back to produce 3D tracing results. Preliminary test results show that the trained neuron tracer is capable of accurately tracing various types of neurons in noisy images.

Pages: 5 to 10

Copyright: Copyright (c) IARIA, 2016

Publication date: November 13, 2016

Published in: conference

ISSN: 2519-8653

ISBN: 978-1-61208-526-5

Location: Barcelona, Spain

Dates: from November 13, 2016 to November 17, 2016