A standard area diagram for Fusarium yellows rating in sugar beet (Beta vulgaris L.)

Abstract

ABSTRACT Members of the Fusarium oxysporum species complex are pathogens of sugar beet causing Fusarium yellows. Fusarium yellows can reduce plant stand, yield, and extractable sugar. Improving host plant resistance against Fusarium -induced diseases, like Fusarium yellows, represents an important long-term breeding target in sugar beet breeding programs. Current methods for rating Fusarium yellows disease severity rely on an ordinal scale, which limits precision for intermediate phenotypes. In this study, we aimed to improve the accuracy and precision of rating Fusarium yellows by developing a standard area diagram (SAD). Two SAD versions were created using images of sugar beets infected with Fusarium oxysporum strain F19. Each version was tested using inexperienced raters. Comparing both the pilot and improved version showed no statistical differences in Lin’s Concordance Correlation Coefficient (LCC) values to assess accuracy and precision between the two versions (Cb = 0.99 for both versions, ρ c = 0.97 and 0.96 for version 1 and 2, respectively). In addition, five naïve Bayesian machine learning models which used pixel classification to determine disease score, were tested for congruency to human estimates in version 2. Root mean square error was lowest compared to the “true” values for the unweighted model and a model where necrotic tissue was given a 2x weight (12.4 and 12.6, respectively). The creation of this standard area diagram enables breeding programs to make consistent, accurate disease ratings regardless of personnel’s’ previous experience with Fusarium yellows. Additionally, more iterations of pixel quantification equations may overcome accuracy issues for rating Fusarium yellows.

Plant phenotyping relevance

フザリウム萎黄病の植物症状を対象に、標準面積図と画像ピクセル分類による病害重症度評価法を開発・検証しており、植物フェノタイピング手法が中心である。

Code and data availability

The paper's data availability statement explicitly deposits the authors' scripts, plant images, and excel sheets (including the RGB classifier training data) on a public GitHub repository, which is paper-specific and actionable.