Machine Learning Assisted Image Analysis

Official page

Image analysis is, by definition, the obtention of metrics describing the objects contained in a particular image. In a perfect situation, these descriptors would accurately represent the biological object in the image and we could consider that the difference between the descriptors and the biological truth is negligible. However, in many cases, artefacts might be present in images in such a way that the representation of the biological object is not accurate anymore. These artefact might be due to the conditions in which the images are taken or the object in itself. Mature root systems, for instance, are complex branched structure, composed of thousands of overlapping and crossing linear segments. The crossing and overlapping are likely to impedes the image analysis and create a gap between the descriptors and the data.

Root system image metrics retrieved with an automated image analysis pipeline.

Here we propose to use machine learning techniques (Random Forests) to help analyse complex root system structures. In short, the idea is to train Random Forest models to estimates traits of interest (that are hard to acquire experimentally) based on descriptors (that are easy to acquire, but with limited biological meaning). We first demonstrated the ability of the Random Forest to predict these traits with a simulated image dataset [Lobet et al. 2017], then with experimental images [Atkinson, Lobet et al, 2017]. In both cases, the Random Forest algorithms have been proven to be more efficient than the traditional approaches.

The machine learning pipeline has been bundled into a R Shiny app, PRIMAL, available here:

Screenshot of PRIMAL

Associated papers

Using a structural root system model for an in-depth assessment of root image analysis pipeline | 2017 | Lobet G*, Koevoets I T*, Noll M*, Meyer P, Tocquin P, Pagès L, Périlleux C | View paper

Combining semi-automated image analysis techniques with machine learning algorithms to accelerate large scale genetic studies | 2017 | Atkinson J*, Lobet G*, Noll M, Meyer P, Griffiths M, Wells D | View paper

Associated presentations

Using structural models to validate and improve root image analysis pipelines | 2016 | View on figshare

blog comments powered by Disqus