Superhuman Accuracy on the SNEMI3D Connectomics Challenge

By Kisuk Lee, Jonathan Zung, Peter H. Li, Viren Jain, H. Sebastian Seung. arXiv.org, 2017

321 citations recorded by the source corpus.

For the past decade, convolutional networks have been used for 3D reconstruction of neurons from electron microscopic (EM) brain images. Recent years have seen great improvements in accuracy, as evidenced by submissions to the SNEMI3D benchmark challenge. Here we report the first submission to surpass the estimate of human accuracy provided by the SNEMI3D leaderboard. A variant of 3D U-Net is trained on a primary task of predicting affinities between nearest neighbor voxels, and an auxiliary task of predicting long-range affinities. The training data is augmented by simulated image defects. The nearest neighbor affinities are used to create an oversegmentation, and then supervoxels are greedily agglomerated based on mean affinity. The resulting SNEMI3D score exceeds the estimate of human accuracy by a large margin. While one should be cautious about extrapolating from the SNEMI3D benchmark to real-world accuracy of large-scale neural circuit reconstruction, our result inspires optimism that the goal of full automation may be realizable in the future.

Subfields: Biomedical Engineering

View Paper: https://www.semanticscholar.org/paper/b907537c602b95948da809f7d4aff4bc959d8ba1

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Biomedical Engineering

1. Qi Tian
2. Dorin Comaniciu
3. Chao Xu
4. Daguang Xu
5. Liang Zheng
6. Yi Yang
7. Zhedong Zheng

Citations in the corpus

• 3D Anisotropic Hybrid Network: Transferring Convolutional Features from 2D Images to 3D Anisotropic Volumes (2017, methodology): “Two more recent studies [17, 21, 11] use small kernels to simulate the large anisotropic kernels.”
• Frequency Domain Compact 3D Convolutional Neural Networks (2020, methodology): “We implement experiments on the Brats 2018 dataset using the Residual 3D U-Net [18] as the backbone.”
• High-precision automated reconstruction of neurons with flood-filling networks (2017, unknown): “State-of-the-art automated neurite reconstruction typically starts with a convolutional network that infers the likelihood of each image location belonging to a neurite, using the intensities of the voxels at and near that location 14–18 .”