Morphological Error Detection in 3D Segmentations
This addresses merge errors in connectomics segmentation, which are critical for accurate biological analysis, but the approach appears incremental as it adapts existing 3D ConvNets to a specific domain problem.
The paper tackled the problem of erroneously merged objects in 3D segmentations for connectomics, which arise from deep learning algorithms relying on localized classification, and showed that MergeNet, a 3D ConvNet trained unsupervised, can detect these errors by learning high-level neuronal morphology, achieving performance across datasets including connectomics data and merged MNIST images.
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.