Skip to main content

Table 1 Quantification of the diseased tissues using the expert-, the model- and the probability-based thresholding approaches

From: High throughput quantitative phenotyping of plant resistance using chlorophyll fluorescence image analysis

   1 4 5 6 7 8 9 10 11
Expert-based thresholding Necrotic tissues (%) 0,02 0,00 0,00 0,02 0,03 0,08 0,39 1,46, 3,44
Wilted tissues (%) 0,06 0,01 0,01 0,06 0,19 0,49 1,02 2,15 4,41
Impacted tissues (%) 0,16 0,05 0,06 0,13 0,20 0,37 0,90 1,74 2,56
Total diseased tissues (%) 0,24 0,06 0,08 0,22 0,42 0,93 2,30 5,35 10,41
Model-based thresholding Total diseased tissues (%) 0,00 5,26 5,26 23,87 42,96 5,64 35,62 49,01 30,39
Probability-based thresholding Strong alteration (%) 0,01 0,00 0,00 0,00 0,00 0,00 0,00 1,35 2,70
Moderate alteration (%) 0,08 0,01 0,03 0,11 0,28 0,70 1,85 2,52 6,67
Weak alteration (%) 0,00 0,95 0,04 0,14 0,45 0,66 1,74 2,96 7,27
  Total diseased tissues (%) 0,09 0,95 0,07 0,25 0,73 1,36 3,59 6,83 16,64
  1. Expert-based thresholding consists in defining Fv/Fm thresholds that enable the selection of areas on Fv/Fm images that match the various stages of the symptom development as observed by trained raters on conventional color images. Healthy, necrotic, wilted, or tissues impacted by the pathogen can be quantified.
  2. Model-based thresholding consists in modeling the pixel-wise Fv/Fm-distributions extracted from each image by mixtures of Gaussian distributions. Such modeling results in the definition of clusters of pixels that correspond to various stages of the alteration of plant tissues. This step is based on the sole analysis of pixel-wise Fv/Fm-distributions and not on the visual observation of symptoms on conventional color images. Therefore, we use the terminology of strong, moderate and weak alteration, to emphasize that this classification is not based on visual observations, and does not necessarily correspond to the various stages of the symptom development as observed by trained raters. When applied directly without preliminary delimitation of the total diseased area, such a modeling over weights artifacts that can occur on healthy tissues, which results in a large overestimation of the proportion of diseased tissues.
  3. Probability-based thresholds consists in the 500-quantile of the merged pixel-wise Fv/Fm-distributions of mock-inoculated samples. Each day of the experiment, the Fv/Fm probability-based threshold allows the splitting of pixels corresponding to healthy and diseased areas. Then whithin the diseased area only, pixel-wise Fv/Fm-distributions are modeled as mixtures of Gaussian distributions to quantify various stages of alteration of plant tissues.