Median-based robust algorithms for tracing neurons from noisy confocal microscope images.

Area of Science:

• Neuroscience
• Image Analysis
• Computational Biology

Background:

• Automated neuron tracing is crucial for analyzing neuronal structures from microscopy images.
• Existing tracing algorithms struggle with low-quality data, such as noisy or discontinuous neuronal structures.
• Such data is common in studies of neuronal growth on microfabricated surfaces.

Purpose of the Study:

• To develop an improved method for fully automatic neuron tracing from noisy confocal microscope images.
• To enhance the precision and recall of tracing algorithms when dealing with low-quality neuronal data.

Main Methods:

• A novel approach was developed by partitioning correlation kernels into subkernels and using their median response for guidance.
• This method refines existing exploratory tracing (vectorization) algorithms.
• Quantitative validation was performed against manual tracing using combined precision and recall metrics.

Main Results:

• Tracing precision improved from 41% to 89% for low-quality data.
• Recall improved by 5%, with better handling of artifacts like discontinuities and hollowness.
• The method generates labeled somas and editable graph-theoretic representations of neuronal structures.

Conclusions:

• The enhanced tracing method significantly improves accuracy and robustness for noisy neuronal images.
• The algorithm is efficient, accurate, and fully automated, making it suitable for large-scale applications.
• This technique is valuable for high-throughput pharmaceutical assays and studies of neuron growth on micro/nano-fabricated structures.