Rshiny expert image comparison app

A Shiny web interface for expert image comparison

I participated to a scientific publication on the analysis of plant cell images. Part of the analysis was to define observation for each cell by visual expertise. The problem is that being able to observe the entire tissue when supposed to give an expertise for each cell biased the cell-centered expertise.
I produce separated images of each cell out of the tissue. I mixed the images from different plant lines submitted to different treatments. I then proposed a R-shiny web interface to randomly show images to the experts and save their observations. This allowed for a non-biased expert image comparison, as the expert had no information on the origin of the cell shown.

A download-upload feature to save and continue analysis on

The expert image comparison was long: 40min for the small analysis and 3h for the complete one. As the shinyapp is freely hosted on the Rstudio servers, it was not possible to save outputs of a specific session on the server. Thus, the web application has been built such that the experts can download a zip file of their partial expertise and come back later. They were then able to upload the beginning of their expertise and continue their analysis when they wanted.  The download/upload feature was a good alternative to the limit of the free hosting service.

You can try this Shinyapp here :
* Code may be available on request.

You can participate !

The R-shiny interface is made such that you can provide your own expertise and retrieve your results. If you want to test your eye against the eye of the co-authors, it is possible ! Data presented in the web interface are the exact image dataset of the publication.
If you provide a complete analysis and send us your results, we may include them in the figures of the R-shiny web interface. The complete analysis may require ~3h, but the special case may only require ~40min.

Rshiny - expert image comparison app
Snapshot of the expert image comparison shinyapp

Quantitative cell micromechanics in Arabidopsis

… or “How to use geostatistical indices to compare image fluorescence of plant cells? …”


Louveaux, M., Rochette, S., Beauzamy, L., Boudaoud, A. and Hamant, O. (2016), The impact of mechanical compression on cortical microtubules in Arabidopsis: a quantitative pipeline. Plant J. Accepted Author Manuscript. doi:10.1111/tpj.13290


Exogenous mechanical perturbations on living tissues are commonly used to investigate whether cell effectors can respond to mechanical cues. However, in most of these experiments, the applied mechanical stress and/or the biological response are described only qualitatively. We developed a quantitative pipeline based on microindentation and image analysis to investigate the impact of a controlled and prolonged compression on microtubule behaviour in the Arabidopsis shoot apical meristem, using microtubule fluorescent marker lines. We found that a compressive stress, in the order of magnitude of turgor pressure, induced apparent microtubule bundling. Importantly, that response could be reversed several hours after the release of compression. Next, we tested the contribution of microtubule severing to compression-induced bundling: microtubule bundling seemed less pronounced in the katanin mutant, in which microtubule severing is dramatically reduced. Conversely, some microtubule bundles could still be observed 16 hours after the release of compression in the spiral2 mutant, in which severing rate is instead increased. To quantify the impact of mechanical stress on anisotropy and orientation of microtubule arrays, we used the nematic tensor based FibrilTool ImageJ/Fiji plugin. To assess the degree of apparent bundling of the network, we developed several methods, some of which were borrowed from geostatistics. The final microtubule bundling response could notably be related to tissue growth velocity that was recorded by the indenter during compression. Because both input and output are quantified, this pipeline is an initial step towards correlating more precisely the cytoskeleton response to mechanical stress in living tissues.

Cell fluorescence converted into a 3D landscape (click on the figure to get the R-code)