Related Experiment Video
Updated: Aug 2, 2026
12:49
A Method for 3D Reconstruction and Virtual Reality Analysis of Glial and Neuronal Cells
Published on: September 28, 2019
12.8K
3D U-Net Segmentation Improves Root System Reconstruction from 3D MRI Images in Automated and Manual Virtual Reality
Tobias Selzner1, Jannis Horn2, Magdalena Landl1
1Forschungszentrum Juelich GmbH, Agrosphere (IBG-3), Juelich, Germany.
Plant Phenomics (Washington, D.C.)
|July 31, 2023
Summary
A novel 2-step workflow using 3D U-Net segmentation significantly improves magnetic resonance imaging (MRI) analysis of root system architecture (RSA). This enhances both manual and automated root reconstruction from soil images.
Area of Science:
- Plant Science
- Imaging Technology
- Computational Biology
Background:
- Magnetic resonance imaging (MRI) is crucial for visualizing root systems in opaque soil.
- Automated reconstruction of root system architecture (RSA) from 3D MRI data is challenging due to low resolution and poor contrast-to-noise ratios (CNRs).
- Manual reconstruction remains the prevalent method despite its labor-intensive nature.
Purpose of the Study:
- To evaluate a novel two-step workflow for automated RSA reconstruction from 3D MRI data.
- To assess the impact of 3D U-Net segmentation on the efficiency and accuracy of manual and automated RSA reconstruction.
- To compare automated reconstructions with manual reconstructions using a virtual reality system.
Main Methods:
- A 3D U-Net model was employed for super-resolution segmentation of MRI images, differentiating roots from soil.
- An automated tracing algorithm was utilized to reconstruct root systems from segmented images.
- Manual reconstructions were performed on both original and segmented MRI images for comparison.
Main Results:
- U-Net segmentation substantially increased manual reconstruction speed by 97% for low CNR and 27% for high CNR images.
- Reconstructed root lengths increased by 20% and 3% after segmentation for low and high CNR images, respectively.
- While automated tracing yielded shorter root lengths, segmentation enabled processing of previously unusable MRI data, and functional root traits showed similar hydraulic behavior across methods.
Conclusions:
- 3D U-Net segmentation is proposed as a key preprocessing step to enhance manual RSA reconstruction workflows.
- The developed automated workflow, despite limitations in root length accuracy, allows for the analysis of challenging MRI datasets.
- Future research will focus on hybrid workflows combining automated scaffolds with manual corrections for improved RSA analysis.
