Characterizing Stomatal and Epidermal Traits Using Peels, Clearing, and AI-Based Image Analysis.

Area of Science:

• Plant anatomy and morphology
• Computational biology
• Ecology

Background:

• Accurate characterization of stomatal and epidermal traits is crucial for understanding plant physiology and ecological adaptation.
• Traditional methods for trait quantification can be labor-intensive and subjective.
• Advancements in image analysis offer potential for high-throughput phenotyping.

Purpose of the Study:

• To present and compare two complementary techniques (epidermal peels and leaf clearing) for assessing stomatal and epidermal traits.
• To evaluate the accuracy and suitability of manual and AI-assisted image analysis methods for trait quantification.
• To assess the performance of different AI segmentation tools (YOLOv8-based, generalist deep learning, pixel-based classifier) across techniques and species.

Main Methods:

• Comparative protocol using epidermal peels and leaf clearing on Fraxinus excelsior and Taraxacum officinale.
• Manual trait quantification via cost-effective tissue clearing and peeling.
• AI-assisted image analysis using YOLOv8 (StoManager1), Cellpose, and a pixel-based classifier.
• Evaluation of segmentation tool accuracy and suitability for stomatal and epidermal trait detection.

Main Results:

• Significant differences in stomatal and epidermal traits were observed between peeling and clearing techniques, consistent across manual and AI analyses.
• AI tools, particularly YOLOv8 (StoManager1) for stomata and Cellpose for epidermal cells, enabled reproducible quantification.
• AI-based trait detection showed high consistency in T. officinale but limitations in F. excelsior, with results diverging from manual measurements.
• Species-specific anatomy and image quality influenced the suitability and accuracy of AI tools.

Conclusions:

• AI-assisted image analysis provides a valuable framework for quantifying stomatal and epidermal traits, enhancing reproducibility.
• The effectiveness of AI tools is influenced by plant species' anatomical features and image quality.
• Expert validation and pilot testing are essential before implementing AI-assisted methods for broader or high-throughput applications in plant phenotyping.