Root segmentation beyond species boundaries: A generalizable framework for anatomical analysis.

Area of Science:

• Plant Biology
• Computational Biology
• Agricultural Science

Background:

• Root anatomical features are crucial for plant performance but are difficult to phenotype due to the high annotation burden of cellular segmentation.
• Current methods for anatomical phenotyping are time-consuming and labor-intensive, limiting scalability across diverse plant species.

Purpose of the Study:

• To develop an efficient and accurate two-stage segmentation framework for plant root anatomical phenotyping.
• To reduce annotation requirements and enable rapid adaptation to new plant species.
• To accelerate the development of climate-resilient crop varieties through scalable root analysis.

Main Methods:

• A two-stage segmentation framework was developed, decomposing multi-class segmentation into species-agnostic tissue identification and subsequent tissue type classification.
• The framework utilizes robust input representations invariant to imaging artifacts and morphological variations.
• The first stage automatically generates tissue boundaries, simplifying manual annotation to tissue labeling.

Main Results:

• The framework achieved state-of-the-art performance on pearl millet and sorghum root cross-sections under various imaging protocols.
• Rapid adaptation to new species was demonstrated with fewer than 40 labeled images.
• Significant reduction in deployment time for root phenotyping was achieved.

Conclusions:

• The developed framework dramatically reduces annotation burden and deployment time for root anatomical phenotyping.
• This efficiency breakthrough enables scalable phenotyping across diverse crop species.
• The method accelerates the development of climate-resilient crop varieties, contributing to global food security.