Optree: a learning-based adaptive watershed algorithm for neuron segmentation.

Area of Science:

• Neuroscience
• Computer Vision
• Biomedical Imaging

Background:

• Accurate segmentation of neuron structures in electron microscopy (EM) images is crucial for understanding neural circuits.
• Existing automated methods often struggle with complex neuronal morphologies and require significant manual correction.

Purpose of the Study:

• To develop a novel, efficient, and accurate algorithm for automatic and interactive neuron segmentation from EM data.
• To improve upon state-of-the-art segmentation techniques by leveraging a tree-structured graphical model.

Main Methods:

• A conditional random field (CRF) model was constructed with the watershed merging tree as its underlying graph.
• The segmentation was obtained via Maximum A Posteriori (MAP) inference on the CRF.
• An interactive framework was developed, utilizing model marginals to identify uncertain regions for user proofreading and subsequent global segmentation refinement.

Main Results:

• The proposed algorithm demonstrated superior performance compared to existing state-of-the-art methods.
• Both training and inference were highly efficient due to the tree-structured nature of the graph.
• The interactive framework effectively incorporated user feedback to globally enhance segmentation accuracy.

Conclusions:

• The developed algorithm offers an efficient and accurate solution for neuron segmentation in EM images.
• The interactive approach allows for targeted user correction, leading to improved segmentation quality.
• This method holds promise for advancing large-scale connectomics research.