3D Neuron Reconstruction in Tangled Neuronal Image With Deep Networks

Area of Science:

• Neuroscience
• Computer Vision
• Biomedical Imaging

Background:

• Accurate 3D neuron reconstruction is crucial for understanding brain function.
• Existing automatic tracing algorithms struggle with neuronal images containing dense nerve fibers.

Purpose of the Study:

• To develop a robust method for segmenting individual neurons from tangled nerve fibers in 3D images.
• To improve the accuracy of 3D neuron tracing and reconstruction.

Main Methods:

• Proposed a novel 3D U-Net Plus network for neuron segmentation.
• Created a synthetic dataset (SYNTANEI) by fusing neurons with nerve fibers from BigNeuron images.
• Utilized dropout, atrous convolution, and Atrous Spatial Pyramid Pooling (ASPP) in the network architecture.

Main Results:

• The 3D U-Net Plus network demonstrated promising segmentation performance on both synthetic and real tangled neuronal images.
• Neuron reconstructions using segmentation results showed significantly improved accuracy compared to reconstructions from original images.
• The method effectively separates neurons from interfering nerve fibers.

Conclusions:

• The proposed 3D U-Net Plus network offers a robust solution for segmenting neurons in complex 3D brain imaging data.
• This segmentation approach enhances the reliability and accuracy of 3D neuron tracing and subsequent analysis.
• The SYNTANEI dataset provides a valuable resource for training and evaluating segmentation models in challenging neuroimaging scenarios.