Abstract

Abstract X-ray microcomputed tomography (μCT) is an invaluable tool for visualizing plant root systems within their natural soil environment noninvasively. However, variations in the x-ray attenuation values of root material and the overlap in attenuation values between roots and soil caused by water and organic materials represent major challenges to data recovery. We report the development of automatic root segmentation methods and software that view μCT data as a sequence of images through which root objects appear to move as the x-y cross sections are traversed along the z axis of the image stack. Previous approaches have employed significant levels of user interaction and/or fixed criteria to distinguish root and nonroot material. RooTrak exploits multiple, local models of root appearance, each built while tracking a specific segment, to identify new root material. It requires minimal user interaction and is able to adapt to changing root density estimates. The model-guided search for root material arising from the adoption of a visual-tracking framework makes RooTrak less sensitive to the natural ambiguity of x-ray attenuation data. We demonstrate the utility of RooTrak using μCT scans of maize (Zea mays), wheat (Triticum aestivum), and tomato (Solanum lycopersicum) grown in a range of contrasting soil textures. Our results demonstrate that RooTrak can successfully extract a range of root architectures from the surrounding soil and promises to facilitate future root phenotyping efforts.