In controlled laboratory tests using synthetic root datasets, participants employing VRoot achieved significantly higher F1 scores, particularly under noisy conditions, and reported improved user experience.
Plant root systems play a critical role in water uptake, nutrient acquisition, and stress resilience. Understanding plant root system architecture (RSA) is essential for advancing sustainable agriculture, yet extracting accurate root data from 3D imaging remains challenging—especially when automatic methods struggle with noisy or complex datasets. Advances in non-invasive 3D imaging—such as magnetic resonance imaging (MRI)—allow researchers to visualize roots in soil without disrupting growth, but converting image data into usable digital models remains a major bottleneck.
Automatic extraction tools often misinterpret root structures due to imaging noise, soil artifacts, or incomplete visibility. Manual corrections by experts can improve results, but conventional desktop interfaces limit spatial perception and interaction efficiency. Based on these challenges, there is a need for more intuitive, accurate, and adaptable tools for 3D root reconstruction.
A study (DOI: 10.1016/j.plaphe.2025.100013) (1) published in Plant Phenomics on 26 March 2025 by Dirk N. Baker’s team Forschungszentrum Jülich GmbH, demonstrates VR’s potential to bridge gaps in root phenotyping, opening new possibilities for plant science and crop improvement.
The research team designed VRoot, a VR application for immersive RSA reconstruction, enabling users to interact directly with 3D volumetric root data. Wearing a head-mounted display and using tracked controllers, participants navigated MRI-derived soil column models, tracing root paths and making real-time adjustments in a fully three-dimensional workspace.
The study compared VRoot against NMRooting, a leading desktop-based root extraction tool, using synthetic datasets with and without water-induced noise. Untrained participants completed root tracing tasks under both conditions.
Results showed that VRoot consistently delivered higher F1 scores—indicating better precision and recall—than NMRooting, with the performance gap widening when water noise was present. VR users also achieved more accurate measurements of root length and inter-lateral distance, and their reconstructions more closely matched ground truth models. Usability assessments, based on the System Usability Scale and pragmatic quality ratings, further favored VRoot, especially for noisy datasets. While some participants initially experienced depth-perception or navigation challenges, brief training sessions mitigated these issues. The findings underscore VR’s ability to enhance spatial awareness, reduce tracing errors, and improve user satisfaction in root phenotyping tasks.
By enabling more accurate manual reconstructions in challenging datasets, VRoot expands the scope of root phenotyping to include diverse soil types and moisture conditions that confound automated algorithms. This capability is valuable for functional-structural plant modeling, where precise RSA data inform simulations of water and nutrient uptake.
The technology could benefit plant breeders, crop scientists, and soil ecologists seeking to link root traits with performance under field-relevant stresses.
Reference:
1. https://doi.org/10.1016/j.plaphe.2025.100013
(Newswise/SG)