Network feature-based phenotyping of leaf venation robustly reconstructs the latent space

Area of Science:

• Plant biology
• Computational biology
• Morphometrics

Background:

• Leaf vein architecture exhibits hierarchical patterns within angiosperm clades.
• Functional demands constrain leaf venation networks, influencing their structure and diversity.
• Previous studies used limited morphological measurements for quantifying leaf venation phenotypes.

Purpose of the Study:

• To develop a high-throughput phenotyping workflow for quantifying leaf vein networks.
• To identify leaf venation-specific morphospace patterns and understand morphological diversities and constraints.
• To quantitatively represent topological aspects of leaf venation for functional and structural analyses.

Main Methods:

• Leaf image acquisition using a feasible system.
• Deep neural network-based leaf vein segmentation.
• Network extraction as an undirected graph and calculation of network features.
• Dimensionality reduction to identify morphospace patterns.

Main Results:

• A 90.6% classification accuracy for five species based solely on network features.
• Identification of a one-dimensional morphospace related to vein loopiness.
• Observed one-dimensional distribution patterns align with theoretical predictions of optimal transport efficiency, construction cost, and robustness.

Conclusions:

• Leaf venation patterns are shaped by functional trade-offs.
• The proposed network feature-based method offers a quantitative descriptor for leaf vein topology.
• This approach is promising for analyzing the functional and structural properties of leaf veins.