A flexible computer vision pipeline to quantify complex interactions in cereal-rust pathosystems

Authors: Carlson, Craig

Submitted to: Meeting Abstract
Publication Type: Abstract Only
Publication Acceptance Date: 12/1/2025
Publication Date: 1/8/2026
Citation: Carlson, C.H. 2026. A flexible computer vision pipeline to quantify complex interactions in cereal-rust pathosystems. Meeting Abstract. Poster No. PE61236.

Interpretive Summary:

Technical Abstract: Oat crown rust, caused by Puccinia coronata f. sp. avenae, is the most destructive disease of cultivated oat (Avena spp.) worldwide, causing substantial yield and quality losses. Durable control requires quantitative resistance, supported by phenotyping methods that resolve the biological complexity of host-pathogen interactions. We developed a high-resolution computer vision pipeline to quantify oat crown rust symptoms and pathogen signs from field-collected images. Using a custom fixed focal camera mount, we captured standardized high-resolution leaf images and annotated them into six biologically meaningful classes: healthy tissue, chlorosis, necrosis, laminar streak/fleck, pustule (lesions containing uredinia), and telia. We trained and optimized multiple deep learning architectures for semantic segmentation, extracted pustule morphology metrics, and compared model-derived traits with traditional breeder-based ratings in the Oat Landrace Diversity (OLD) Panel. Image-derived traits revealed novel patterns of host response as well as morphological indicators of quantitative resistance not captured by the breeder’s eye. For instance, this is the first report of near perfect scaling relationships between pustule density and those tied to shape. By providing trait-specific phenotypes that can be integrated into genomic analyses, our pipeline enables a more mechanistic understanding of plant-pathogen interactions by improving the resolution, objectivity, and richness of phenotypic observations.