A shape-based method for automatic and rapid segmentation of roots in soil from X-ray computed tomography images: Rootine

摘要

Aims X-ray computed tomography (CT) is widely recognized as a powerful tool for in-situ quantification of root system architecture (RSA) in soil. However, employing X-ray CT to identify the spatio-temporal dynamics of RSA still remains a challenge due to non-automatic, time-consuming image processing protocols and their poor recovery of fine roots in soil. Methods Here we present a new protocol (Rootine) to segment roots rapidly and precisely down to fine roots with two voxels in diameter (90 mu m in pots with 70 mm in diameter). This is facilitated by feature detection of the tubular shape of roots, an approach that was originally developed for detecting blood vessels in medical imaging. Results In comparison to established root segmentation methods, Rootine produced a more accurate root network, i.e. more roots and less over-segmentation. Root length quantified by X-ray CT showed high correlation with results by root washing combined with 2D light scanning (R-2 = 0.92). Tests with different soil materials showed that the recovery of roots depends on signal-to-noise ratio but can be up to 99% for a favorable contrast between fine roots and background. Conclusions This new protocol provides great efficiency to study RSA in undisturbed soil. As it is fully automated it has the potential for high-throughput root phenotyping and related modelling.