Stomata morphology measurement with interactive machine learning: accuracy, speed, and biological relevance?

Area of Science:

• Plant Science
• Computational Biology
• Genetics

Background:

• Stomatal morphology is crucial for plant gas exchange, water use efficiency, and ecological adaptation.
• Traditional manual measurement of stomatal traits is time-consuming and labor-intensive.
• Machine learning (ML) offers a potential solution for high-throughput phenotyping.

Purpose of the Study:

• To evaluate a U-Net based interactive ML software with corrective annotation for stomatal morphology phenotyping.
• To assess the efficiency and accuracy of the ML approach across diverse plant image datasets.
• To determine the feasibility of a single ML model for analyzing varied stomatal data.

Main Methods:

• Training a single U-Net model on five diverse stomatal image datasets.
• Testing the model's performance on unseen data for stomatal density and size.
• Applying thresholding techniques to U-Net segmentations to enhance accuracy.
• Comparing the speed and accuracy of semi-automatic ML annotation with manual methods.

Main Results:

• High accuracy achieved for stomatal density (R²=0.98) and size (R²=0.90).
• Thresholding improved accuracy, especially for density measurements.
• Semi-automatic ML annotation was five times faster than manual annotation with comparable accuracy.
• ML metrics like F1 score correlate with statistical analysis accuracy.

Conclusions:

• Interactive ML with corrective annotation is a robust and accessible tool for plant phenotyping.
• The developed ML approach significantly accelerates stomatal trait analysis, reducing technical barriers.
• The model enables detection of significant biological differences in stomatal morphology across various conditions.