Dendritic tree extraction from noisy maximum intensity projection images in C. elegans

Area of Science:

• Neuroscience
• Computational Biology
• Image Analysis

Background:

• Neuronal dendritic tree morphology is crucial for understanding mechanosensory function in C. elegans.
• Extracting dendritic trees from Maximum Intensity Projections (MIPs) is challenging due to noise and reliance on manual methods.
• Automated and reliable 2D segmentation of dendritic trees is needed.

Purpose of the Study:

• To develop an automated, reliable 2D segmentation method for dendritic trees in noisy MIPs.
• To utilize a statistical learning framework for improved dendritic tree extraction (DTE).
• To provide a faster and more accurate alternative to manual segmentation.

Main Methods:

• Employs a statistical learning framework using limited labeled training data on MIPs.
• Learns noise models of texture-based features for tree structures and background.
• Uses probabilistic/Bayesian inference with noise models, followed by morphological thinning for skeletonization.

Main Results:

• Demonstrates significant improvements in tree-structure segmentation over traditional intensity-based methods via Leave-One-Out Cross Validation (LOOCV).
• Achieves qualitative and quantitative improvements across various imaging conditions, validated by Receiver Operator Characteristic (ROC) curves.
• Successfully extracts skeletonized structures from novel MIP samples, outperforming state-of-the-art tracing software.

Conclusions:

• The developed DTE method provides robust dendritic tree segmentation in noisy MIPs.
• Outperforms traditional intensity-based methods, offering a reliable framework for dendritic tree identification.
• Accelerates morphological characterization of neuronal arborization.