Semantic segmentation of microscopic neuroanatomical data by combining topological priors with encoder-decoder deep...

Area of Science:

• Neuroscience
• Computational Biology
• Machine Learning

Background:

• Traditional neuron tracing is labor-intensive and infeasible for large-scale brain imaging data.
• Existing deep learning methods require extensive annotated data and high error rates, necessitating human review.

Purpose of the Study:

• To develop an efficient and accurate automated method for neuronal tracing and connectivity analysis in large-scale brain images.
• To integrate topological data analysis with deep learning to overcome limitations of current approaches.

Main Methods:

• A hybrid architecture combining discrete Morse theory (topological data analysis) with deep neural networks was developed.
• The architecture was adapted into a high-performance pipeline for semantic segmentation of whole-brain image data.

Main Results:

• The hybrid architecture achieved near 90% precision/recall in detecting neuronal structures like connectivity and synaptic swellings.
• Performance significantly surpassed previous methods, approaching human observer accuracy.
• The pipeline successfully segmented neuronal compartments in light microscopic whole-brain images.

Conclusions:

• The hybrid approach offers a significant advancement in automated neuronal tracing and connectivity analysis.
• Incorporating discrete Morse techniques into deep nets enhances accuracy and efficiency for large-scale neuroscience data.
• This method shows potential for generalization to other complex biological data domains.