Plant roots are essential to the structure and function of plants. They provide access to belowground water and nutrients [1
], facilitate anchorage of plants in soils [4
], mediate chemical defense belowground [6
], and serve as sites of important symbioses with microbiota [8
]. However, comparatively little is known about the architecture of plant root systems and their relationship to overall plant function given the relative inaccessibility of belowground tissue [10
]. Recently, standardized field procedures have been developed for quantifying RSA using excavated and washed roots [11
]. In addition, a number of non-destructive approaches to imaging and quantifying RSA have been developed. The approaches include X-ray computed tomography [12
], nuclear magnetic resonance (NMR) microscopy [16
], magnetic resonance imaging [17
], laser scanning [19
], as well as a growing number of imaging methods using plants grown in transparent media [20
]. These methods are part of a rapidly growing field of “plant phenomics” whose objective is to link plant genotypes to plant phenotypes [26
], particularly in the service of plant biotechnology and crop improvement [30
Analyzing large numbers of plant root system images requires the use of software tools specifically designed for the high-throughput estimation of RSA traits. A number of different tools are available for the analysis of RSA. For example, some programs are specialized for the analysis of images from a specific apparatus, e.g., images from minirhizotrons [32
]. Next, general tools are available for in-depth analysis of individual monocot root systems regardless of apparatus. These tools rely on significant user input for processing although they can be used in a batch mode [22
]. Similarly, general purpose image processing programs such as ImageJ may be flexible enough to perform many specialized tasks [38
]. In practice, end-users often utilize point-and-click approaches that are not scaleable to large numbers of edges within a root system (for an exception, see the recently released software “SmartRoot” which is designed for semi-automated analysis of the hierarchical structure of root systems [39
]). Finally, software is also available that can accurately characterize the simpler dicot root system of Arabidopsis
], in a high-throughput fashion [40
], but that is not yet suitable for studying the intricate monocot root systems of rice and maize.
GiA Roots is a software tool that estimates RSA traits from a large number of root system images. The main distinguishing characteristics of GiA Roots are that it is both specifically designed for high-throughput analysis and it is extensible. A typical user begins by interactively setting parameters to enable GiA Roots to identify roots from the background, i.e., segmenting the image. Next, the user selects traits of interest for measurement. Finally, the user instructs GiA Roots to automatically estimate traits for images. The GiA Roots pipeline of steps can be executed from the GUI or command line tool as part of a fully automated workflow. A technical user can add new traits to the GiA Roots pipeline using the application programming interface of GiA Roots (GiA-API).
In this manuscript, we describe the implementation of the software, including the main computational steps. We discuss the use of GiA Roots, including validating its measurements via estimation of 19 RSA traits on a large-scale data set of 2393 images from rice root systems. We demonstrate the segmentation capabilities of GiA Roots on root system images. We also demonstrate how GiA Roots can be extended to include a novel trait presented in a recent RSA analysis of 3D reconstructions of root networks [22
] but not contained in our earlier study of 2D images of root networks [21
]. Finally, we discuss the limitations of the software and plans for further development.