Abstract
This article describes an immersive extended reality (XR) reconstruction tool for root system architectures (RSA) from 3D volumetric scans of soil columns. We conducted a laboratory user study to assess the performance of new users with our software in comparison to established desktop software. Utilizing a functional-structural plant model (FSPM) to derive synthetic root architectures, we objectively quantified reconstruction accuracy. Our results demonstrate an increase in RSA reconstruction accuracy ($F_1$ score) compared to state-of-the-art desktop software and more robust extraction quality. This work highlights the advantages of immersive techniques for 3D data extraction in plant science.
Introduction
Understanding Root System Architecture (RSA)—the spatial configuration of a plant's roots—is vital for sustainable agriculture, as it dictates how plants absorb water and nutrients. However, roots are hidden underground. While non-invasive 3D imaging like MRI can "see" through soil, the resulting data is often "noisy" (blurry or cluttered) due to soil composition and water content.
Current software typically relies on 2D desktop interfaces where users must navigate 3D volumes on a flat monitor. This makes it difficult to perceive depth, distinguish between overlapping roots, or follow a single root through gaps in the data. Fully automated algorithms often fail in these complex scenarios, leaving researchers with a bottleneck: the need for a more intuitive way for humans to manually trace or correct 3D skeletons.
The VRoot System
To address these challenges, we developed VRoot, an XR application built using the Visualization ToolKit (VTK) for the HTC Vive Pro. VRoot moves the reconstruction task from a 2D screen into an immersive 3D environment.
Key technical features include:
- Immersive Visualization: Users view the root system as a 3D volume they can physically walk around or rotate.
- Dithered Translucency: A specialized rendering technique that helps users interpret depth and distance within the semi-transparent "cloud" of MRI data.
- Two-Handed Interaction: Users utilize one controller to navigate or "grab" the volume and the other to place nodes and "draw" the root architecture.
Methodology and Validation
To objectively measure accuracy, the study utilized a synthetic dataset generated via the CPlantBox modeling framework. This provided a "ground truth" to compare against manual traces.
- Datasets: Simulated bean root systems were converted into synthetic MRI volumes under two conditions: "No Water" (clean signal) and "Water+" (high noise, simulating soil moisture interference).
- Comparative Study: VRoot was compared against NMRooting, a state-of-the-art desktop application, in a study involving 20 participants. The primary benchmark was the $F_1$ score, balancing precision and recall.
Key Results and Insights
The study revealed several advantages of the immersive approach:
- Stereoscopic Advantage: 3D immersion solves the "occlusion problem." In VR, users can simply lean to see the physical gap between crossing roots that appear as a single point on a 2D screen.
- Robustness to Noise: VR users were significantly more resilient to data noise. In the "Water+" condition, desktop accuracy plummeted, while VR users maintained high $F_1$ scores.
- Precision in Traits: The relative error for measuring inter-lateral distance was 0.234 for VR versus 0.617 for desktop—a nearly 3x improvement.
- Usability: VRoot achieved a System Usability Scale (SUS) score of 82 ("excellent"), compared to 67 ("marginal") for the desktop software. NASA-TLX scores indicated that VRoot was not significantly more taxing than traditional input methods.
Conclusion
While AI continues to improve, this research argues that for complex biological data, providing humans with better "Immersive Analytics" tools is currently more effective than relying on fully autonomous systems. Future work will explore "hybrid" approaches, using AI for initial tracing and VRoot for rapid human verification and correction, combining machine speed with human spatial reasoning.
Summary updated