Automatic 3D Single Neuron Reconstruction With Exhaustive Tracing

Zihao Tang, Donghao Zhang, Siqi Liu, Yang Song, Hanchuan Peng, Weidong Cai; Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2017, pp. 126-133

The digital reconstruction of neuronal morphology from single neurons, also called neuron tracing, is a crucial process to gain a better understanding of the relationship and connections in neuronal networks. However, the fully automation of neuron tracing remains a big challenge due to the biological diversity of the neuronal morphology, varying image qualities captured by different microscopes and large-scale nature of neuron image datasets. A common phenomenon in the low quality neuron images is the broken structures. To tackle this problem, we propose a novel automatic 3D neuron reconstruction framework named exhaustive tracing including distance transform, optimally oriented flux filter, fast-marching and hierarchical pruning. The proposed exhaustive tracing algorithm shows a robust capability of striding over large gaps in the low quality neuron images. It outperforms state-of-the-art neuron tracing algorithms by evaluating the tracing results on the large-scale First-2000 dataset and Gold dataset.