Accelerated Haustoria Segmentation Enables Rapid Gene Function Analysis in Cereal-Powdery Mildew Pathosystems.

Abstract

Reliable, high-throughput quantification of early fungal infection events is crucial for gene-function studies, but it remains labor-intensive. We report an open-source pipeline that automates the detection of β-glucuronidase (GUS-stained) epidermal cells and the intracellular haustoria formed by powdery mildew on barley and wheat leaves. Whole-slide images are captured with a commercial scanner, focus-projected, tiled, and analyzed by deep-learning models trained on expertly annotated datasets. A You Only Look Once (YOLO) network identifies GUS-positive cells, while a companion segmentation model pinpoints haustoria within each cell; automatic focus-layer selection preserves fine structural detail. The workflow runs in minutes per slide on a single workstation and maintains near-perfect agreement with manual counts in both barley and wheat, demonstrating robust cross-species transferability. By delivering single-cell readouts with minimal user input, the pipeline enables rapid functional validation screens and supports large-scale phenotyping of cereal-powdery mildew interactions.