Fig 1.
Schematic representation of lesion growth monitoring through imaging.
The initial RGB images (a) are first registered to align stipules in time (b). Afterwards, a supervised segmentation is performed to produce probability maps indicating the probability of each pixel to be in either healthy, symptomatic or background classes. Probability images of the symptomatic state (c) are used for fitting the Fisher-KPP model. Images of day 3 are used as initial conditions while the remaining 4 images are used to estimate the pathogen local growth rate and diffusion coefficient
that are actually two distinct life-history traits of within-host pathogen spread.
Fig 2.
Visualization of model prediction against image data.
The solution of the fitted Fisher-KPP equation, i.e. with optimal estimated parameters , is represented through time by contours (0.2, 0.3, 0.4, 0.5) overlying the probability images of the symptomatic class for example stipules of both cultivars, i.e. Solara n°1 (a) and James n°17 (b). This comparison between the spatial model and the data is also available in S1 and S2 Movies.
Fig 3.
Visualization of stipules deformation in time.
Change in the Jaccard index with time for cultivars James (a) and Solara (b). At each time after inoculation the Jaccard index was calculated in comparison with the image at day 3, also used as a reference for image registration.
Fig 4.
Comparison of symptomatic surface predicted by the reaction-diffusion models against the surface obtained after pixel-based image segmentation for all images used to fit the model.
For each cultivar, i.e. James (black points) and Solara (gray points), there are thus 16 stipules × 4 dates points. The black line is the first bisector that indicates a perfect agreement between values while the red line is the estimated linear relationship between prediction and observation considering all data (slope = 1.05, intercept = −3392.0). In detail, the relationship for James cultivar was better (slope = 1.01, intercept = −1418.7) than for James and (slope = 1.14, intercept = −6947.1).
Fig 5.
Distributions of the estimated parameters.
a) diffusion coefficient with a mean values of 0.55 for James against 1.54 for the more susceptible cultivar Solara, b) local growth rate
with mean values of 1.33 and 1.26 for respectively James and Solara. The 32 estimated parameters (16 for each cultivar) are available in S6 Appendix.