A deep learning-integrated micro-CT image analysis pipeline for quantifying rice lodging resistance-related traits

Area of Science:

• Plant Science and Agronomy
• Computational Biology and Bioinformatics
• Agricultural Engineering

Background:

• Lodging in rice significantly reduces yield and harvest efficiency, often linked to suboptimal culm structure.
• Traditional methods for measuring rice culm traits are destructive, time-consuming, and labor-intensive, hindering large-scale breeding.
• Developing robust lodging resistance is crucial for improving rice productivity and ensuring food security.

Purpose of the Study:

• To develop a high-throughput, non-destructive method for quantifying rice culm morphology and lodging resistance.
• To integrate micro-computed tomography (micro-CT) imaging with deep learning (SegNet) for automated trait extraction.
• To validate the accuracy of the developed pipeline against manual measurements and assess its utility in early growth stages.

Main Methods:

• Utilized a high-throughput micro-CT-RGB imaging system for data acquisition.
• Developed a deep learning-based image analysis pipeline (SegNet) to extract 24 rice culm traits.
• Compared automated measurements with manual measurements for key traits (major axis, minor axis, wall thickness) in 104 indica rice accessions.

Main Results:

• The automated pipeline achieved high accuracy, with mean absolute percentage errors below 10% for major axis, minor axis, and wall thickness compared to manual measurements.
• Models predicting bending stress using culm traits showed good accuracy (R² = 0.722 at mature stage, R² = 0.544 at tillering stage), enabling early-stage assessment.
• Quantified relationships between bending stress, biomass, culm density, and drought resistance, revealing trade-offs.

Conclusions:

• A deep learning-integrated micro-CT image analysis pipeline enables rapid (∼4.6 min/plant) and accurate quantification of rice culm phenotypes.
• This non-destructive method effectively assesses lodging resistance, even in early growth stages, facilitating high-throughput screening.
• The pipeline will significantly aid in breeding programs aiming to develop rice varieties with improved lodging resistance and productivity.