Deep Open-Curve Snake for Discriminative 3D Neuron Tracking.

Area of Science:

• Computational Neuroscience
• Machine Learning for Biology
• Neuroimaging Analysis

Background:

• Open-Curve Snake (OCS) is used for 3D neurite tracking but struggles with noisy, weak signals and initial seed sensitivity.
• Existing methods rely heavily on image gradients, limiting performance in real-world applications.

Purpose of the Study:

• To develop a novel deep learning framework, Deep Open-Curve Snake (DOCS), for robust 3D neuron tracking.
• To overcome the limitations of traditional OCS by incorporating learnable components and energy minimization.

Main Methods:

• DOCS employs a discriminative framework that simultaneously learns a 3D distance-regression discriminator and a 3D deeply-learned tracker.
• The tracking process involves convolutional neural network predictions for deformation fields, stretching directions, and local radii, iteratively updated via energy minimization.
• A tractable energy function incorporating fitting forces and curve length guides the iterative updates.

Main Results:

• DOCS demonstrated superior performance on the BigNeuron and Diadem datasets compared to existing neuron tracing methods.
• The framework achieved state-of-the-art results, improving average overlap and distance scores on the BigNeuron dataset by 1.7% and 17%, respectively.
• On the Diadem dataset, DOCS improved the average overlap score by 4.1%.

Conclusions:

• DOCS significantly advances 3D neuron tracking by leveraging deep learning for segmentation, tracing, and reconstruction of complete neuron structures.
• The integrated approach effectively handles noisy and weak filament signals, outperforming traditional methods.
• DOCS offers a robust and accurate solution for analyzing complex neuronal architectures in volumetric data.