Superhuman Accuracy on the SNEMI3D Connectomics Challenge (2017-05-31T00:00:00.000000Z)

TL;DR

This work reports the first submission to surpass the estimate of human accuracy provided by the SNEMI3D leaderboard, and inspires optimism that the goal of full automation may be realizable in the future.

Abstract

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.

Authors

Peter H. Li

2 papers

Viren Jain

2 papers

Kisuk Lee

2 papers

TL;DR

Abstract

Authors

References35 items

DeepEM3D: approaching human‐level performance on 3D anisotropic EM image segmentation

Convolutional Neural Pyramid for Image Processing

Volumetric ConvNets with Mixed Residual Connections for Automated Prostate Segmentation from 3D MR Images

Residual Deconvolutional Networks for Brain Electron Microscopy Image Segmentation

Multicut brings automated neurite segmentation closer to human performance

Feature Pyramid Networks for Object Detection

A Multi-Pass Approach to Large-Scale Connectomics

Deep Convolutional Neural Network Design Patterns

Flood-Filling Networks

3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation

V-Net: Fully Convolutional Neural Networks for Volumetric Medical Image Segmentation

Residual Networks are Exponential Ensembles of Relatively Shallow Networks

Learning to Refine Object Segments

A guide to convolution arithmetic for deep learning

Deep Residual Learning for Image Recognition

Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs)

Crowdsourcing the creation of image segmentation algorithms for connectomics

Saturated Reconstruction of a Volume of Neocortex

U-Net: Convolutional Networks for Biomedical Image Segmentation

Image Segmentation by Size-Dependent Single Linkage Clustering of a Watershed Basin Graph

Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift

Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification

Adam: A Method for Stochastic Optimization

Facial Landmark Detection by Deep Multi-task Learning

Caffe: Convolutional Architecture for Fast Feature Embedding

Graph-based active learning of agglomeration (GALA): a Python library to segment 2D and 3D neuroimages

Toward large-scale connectome reconstructions

Learned versus Hand-Designed Feature Representations for 3d Agglomeration

Machine Learning of Hierarchical Clustering to Segment 2D and 3D Images

Learning to Agglomerate Superpixel Hierarchies

Contour Detection and Hierarchical Image Segmentation

Convolutional Networks Can Learn to Generate Affinity Graphs for Image Segmentation

Supervised Learning of Image Restoration with Convolutional Networks

Comparing clusterings---an information based distance

Natural Scales in Geographical Patterns

Field of Study

Journal Information

Name

Volume

Venue Information

Name

Type

URL

Alternate Names