AlGrow: a graphical interface for easy, fast and accurate area and growth analysis of heterogeneously colored targets.

Abstract

Image analysis is widely used in plant biology to determine growth rates and other phenotypic characters, with segmentation into foreground and background being a primary challenge. Statistical clustering and learning approaches can reduce the need for user input into this process, though these are computationally demanding, can generalise poorly and are not intuitive to end users. As such, simple strategies that rely on the definition of a range of target colors are still frequently adopted. These are limited by the geometries in color space that are implicit to their definition; i.e. thresholds define cuboid volumes and selected colors with a radius define spheroid volumes. A more comprehensive specification of target color is a hull, in color space, enclosing the set of colors in the image foreground. We developed AlGrow, a software tool that allows users to easily define hulls by clicking on the source image or a three-dimensional projection of its colors. We implemented convex hulls and then alpha-hulls, i.e. a limit applied to hull edge length, to support concave surfaces and disjoint color volumes. AlGrow also provides automated annotation by detecting internal circular markers, such as pot margins, and applies relative indexes to support movement. Analysis of publicly available Arabidopsis image series and metadata demonstrated effective automated annotation and mean Dice coefficients of >0.95 following training on only the first and last images in each series. AlGrow provides both graphical and command line interfaces and is released free and open-source with compiled binaries for the major operating systems.