Minimizing Probability Graph Connectivity Cost for Discontinuous Filamentary Structures Tracing in Neuron Image

Area of Science:

• Neuroscience
• Biomedical Imaging
• Computational Biology

Background:

• Neuron tracing is essential for understanding brain function and disease.
• Tracing discontinuous filamentary structures in noisy images presents significant challenges.
• Existing methods struggle with assembling fragmentary traces, leading to topological errors.

Purpose of the Study:

• To develop a precise graph connectivity theoretical method for filamentary structure tracing in neuron images.
• To address the problem of discontinuous and noisy structures in neuronal and medical imaging.
• To improve the accuracy and completeness of neuron tracing.

Main Methods:

• Utilizing a region-to-region based tracing method on Convolutional Neural Network (CNN) predicted probability to build initial subgraphs.
• Employing a dynamic linear programming function to solve the global connection problem of fragmented subgraphs.
• Minimizing graph connectivity cost, calculating breakpoint costs via probability strength and minimum cost paths.

Main Results:

• The proposed method accurately traces discontinuous filamentary structures in challenging neuronal images.
• Experimental results demonstrate superior performance compared to existing tracing methods.
• Achieved results comparable to manual tracing, even in complex discontinuous scenarios.

Conclusions:

• The graph connectivity method offers a precise solution for filamentary structure tracing in neuron images.
• The technique effectively handles noise and fragmentation, improving topological accuracy.
• Demonstrated potential for tracing other tubular objects, such as vessels, in medical images.