Automated Neuron Tracing Using Content-Aware Adaptive Voxel Scooping on CNN Predicted Probability Map

Area of Science:

• Neuroscience
• Computational Biology
• Image Analysis

Background:

• Neuron tracing is vital for analyzing neural circuits and brain function.
• Existing automatic tracing methods struggle with complex, noisy, and fragmented neuronal structures.

Purpose of the Study:

• To develop an accurate and robust method for automatic neuron tracing.
• To overcome limitations of current methods in handling complex and broken neurites.

Main Methods:

• A 3D residual convolutional neural network (CNN) was used for noise suppression and probability map prediction.
• Content-aware adaptive voxel scooping was applied to probability maps for tracing, considering neurite properties like distance, connectivity, and probability continuity.
• Neuron tree graphs were constructed using a length-first criterion.

Main Results:

• The proposed method demonstrated superior performance compared to state-of-the-art techniques, particularly on datasets with complex and broken neurites.
• High accuracy in tracing was achieved on both public BigNeuron and fluorescence micro-optical sectioning tomography (fMOST) datasets.
• The method shows significant potential for large-scale neuron tracing applications.

Conclusions:

• The novel content-aware adaptive voxel scooping method significantly improves automatic neuron tracing accuracy.
• This approach effectively addresses challenges posed by complex neuronal morphologies and image noise.
• The method holds promise for advancing large-scale neural circuit reconstruction and analysis.