Morphological Error Detection in 3D Segmentations

TL;DR

This paper introduces MergeNet, a 3D convolutional neural network that detects morphological merge errors in neuronal segmentation, outperforming traditional localized classification methods by leveraging high-level shape information.

Contribution

The paper presents MergeNet, an unsupervised 3D ConvNet that effectively detects merge errors in connectomics data and generalizes across different datasets, including synthetic MNIST merges.

Findings

MergeNet accurately detects merge errors in connectomics datasets.

It generalizes well across different datasets and types of morphological errors.

The approach outperforms traditional localized classification methods.

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.