Morphological Error Detection in 3D Segmentations
arXiv:1705.10882
Abstract
Deep learning algorithms for connectomics rely upon localized classification, rather than overall morphology. This leads to a high incidence of erroneously merged objects. Humans, by contrast, can easily detect such errors by acquiring intuition for the correct morphology of objects. Biological neurons have complicated and variable shapes, which are challenging to learn, and merge errors take a multitude of different forms. We present an algorithm, MergeNet, that shows 3D ConvNets can, in fact, detect merge errors from high-level neuronal morphology. MergeNet follows unsupervised training and operates across datasets. We demonstrate the performance of MergeNet both on a variety of connectomics data and on a dataset created from merged MNIST images.
13 pages, 6 figures
References in corpus (15)
- TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems
- Semi-Supervised Classification with Graph Convolutional Networks
- Learning Convolutional Neural Networks for Graphs
- Training Convolutional Networks with Noisy Labels
- Towards Biologically Plausible Deep Learning
- Multi-view Convolutional Neural Networks for 3D Shape Recognition
- Combining Fully Convolutional and Recurrent Neural Networks for 3D Biomedical Image Segmentation
- Automatic Neuron Detection in Calcium Imaging Data Using Convolutional Networks
- A Multi-Pass Approach to Large-Scale Connectomics
- Flood-Filling Networks
- A Context-aware Delayed Agglomeration Framework for Electron Microscopy Segmentation
- Recursive Training of 2D-3D Convolutional Networks for Neuronal Boundary Detection
- Automatic Neuron Type Identification by Neurite Localization in the Drosophila Medulla
- Toward Streaming Synapse Detection with Compositional ConvNets
- Large-Scale Electron Microscopy Image Segmentation in Spark