Extracting Metrics for Three-dimensional Root Systems: Volume and Surface Analysis from In-soil X-ray Computed Tomography Data

Summary

A methodology for obtaining visual and quantitative root structure information from X-ray computed tomography data acquired in-soil is presented.

Abstract

Plant roots play a critical role in plant-soil-microbe interactions that occur in the rhizosphere, as well as processes with important implications to climate change and crop management. Quantitative size information on roots in their native environment is invaluable for studying root growth and environmental processes involving plants. X-ray computed tomography (XCT) has been demonstrated to be an effective tool for in situ root scanning and analysis. We aimed to develop a costless and efficient tool that approximates the surface and volume of the root regardless of its shape from three-dimensional (3D) tomography data. The root structure of a Prairie dropseed (Sporobolus heterolepis) specimen was imaged using XCT. The root was reconstructed, and the primary root structure was extracted from the data using a combination of licensed and open-source software. An isosurface polygonal mesh was then created for ease of analysis. We have developed the standalone application imeshJ, generated in MATLAB¹, to calculate root volume and surface area from the mesh. The outputs of imeshJ are surface area (in mm²) and the volume (in mm³). The process, utilizing a unique combination of tools from imaging to quantitative root analysis, is described. A combination of XCT and open-source software proved to be a powerful combination to noninvasively image plant root samples, segment root data, and extract quantitative information from the 3D data. This methodology of processing 3D data should be applicable to other material/sample systems where there is connectivity between components of similar X-ray attenuation and difficulties arise with segmentation.

Introduction

Roots, as part of the rhizosphere^2-5, represent an "invisible" part of plant biology since soil makes it difficult to image roots non-invasively ^{6, 7}. However, studying root growth and interaction within the soil environment is critical to understanding root/plant growth and nutrient cycling, which in turn affect forestation, food security, and climate. X-ray computed tomography (XCT) has proven to be a valuable tool for noninvasive imaging of plant root samples in their local environments ⁸. In order to measure root development and dimensional changes under different conditions, and be able to compare data from different datasets/specimens, one needs to extract quantitative information from the tomography data. Segmentation of the root data from that of the surrounding soil, that is, the isolation of the root image from everything else around it (including, for example, a neighboring plant) is a critical step before accurate size analysis can be done. However, a simple thresholding approach is often unfeasible for root data. The challenges associated with imaging plant roots in soil include variations in the X-ray attenuation properties of the root material, and the overlap in attenuation values between root and soil caused by water and organic matter. These issues have been superbly addressed recently by Mairhofer et al. in their visual tracking tool RooTrak ^{7, 9}. The next step after a successful segmentation is the accurate determination of root volume and surface area. The volume may be estimated by counting the number of voxels and multiplying by the voxels' size cubed as shown before⁷. For a more accurate determination of root surface area and volume, the isosurface of the segmented root system can be represented by a mesh of triangles, using an algorithm known as Marching Cubes ¹⁰. The open-source ImageJ¹¹ can be employed to approximate the root volume based on the Marching Cubes algorithm. To the best of our knowledge, only a limited number of open-source software dedicated to calculating tomography-based volume/surface data for root specimens in the centimeter range and above is currently available ¹². One open-source software we looked at¹³ focuses on root growth and is aimed at cellular features enabling quantitative volume analysis at single-cell resolution. Some open-source software dedicated to whole root systems¹⁴ is excellent for small-diameter tubular root systems based on the approximation that their shape is actually tubular. However, some work with 2D images and are unable to handle 3D stacks¹⁴. Furthermore, the tubular shape approximation may not be valid when root systems with rough surfaces and non-uniform shapes, such as those of trees, are studied. Another approach¹⁵ uses two-dimensional (2D) rotational image sequences innovatively circumventing the need for a costly CT scanner. It measures, records, and displays root system lengths. The software we have tested from those only available commercially^16-18; one does not appear to be able to handle 3D image stacks¹⁶, the second is a leaf area and root length measurement tool¹⁷, while the third is based on color analysis¹⁸. Based on this survey, we suggest that a costless option that approximates the surface and volume of the root regardless of its shape from 3D tomography data is desirable.

Building on the freely available RooTrak and ImageJ, we have developed a program, named imeshJ (see Supplemental Code File) which processes an isosurface mesh (surface stereolithography file) generated from segmented root data, and calculates the volume and surface area of the root by doing simple geometrical calculations on the mesh triangle index data. Here we report a method that combines the use of XCT imaging, data reconstruction and visualization (software CT Pro 3D and VG Studio), segmentation of the root of the specimen from the soil in the 3D data (open-source software ImageJ and RooTrak), and extraction of the surface and volume information from a triangular mesh (ImageJ and the computer code imeshJ).

Protocol

Caution: The operation of an X-ray tomography scanner requires both general radiation training, and instrument-specific radiation safety training. All corresponding procedures relevant to the experimenter's laboratory should be followed. 

1. Root Imaging

Note: This step describes the imaging of a grass specimen held in its original soil in a tubular plastic pot (a plastic tube with a diameter of 40 mm, a height of 210 mm, and wall thickness of about 2 mm).

1. Place potted plant on the sample manipulator of the instrument at a distance desired for target magnification. For a plant in a 2-inch diameter holder, the sample to source distance should be about 3 inches (7 cm).
2. Adjust the X-ray scan settings to achieve optimal color (grey level) contrast in detector image. Note: These settings are available in the instrument control software used.
   1. Set X-ray power settings; 85 kV and 190 μA were used in this example.
   2. Set exposure time. Here, a relatively long exposure time of 1 sec was used for better signal-to-noise ratio.
   3. Set number of projections and frames per projection; 4 frames per projection for a total of 3,142 projections is suggested for good data statistics.
   4. Run a shading correction using the measurement conditions set above by selecting the "Shading correction" tab, and clicking "Create".
      Note: The shading correction compensates for the variation in response of the imaging device's pixels when illuminated with a constant set of X-ray flux. The process takes blank images (with sample removed from beam path) with the X-ray beam turned on, and with the beam turned off. This correction is applied to all images collected.
   5. Select the "Minimize ring artefacts" option (also called "shuttling mode"); the sample will be rotated in angular steps whilst the projection images are acquired. This leads to data acquisition at a slower rate, but helps eliminate ring artefacts.
   6. Start the scan by clicking the "Acquire" button under the Acquisition tab (with the settings outlined above, image collection will take approximately 4 hr).

2. Data Reconstruction 

Note: This section describes the reconstruction of 3D volume data from the raw images (radiographs from the CT scan).

1. Load the raw data into the program.
2. Compare first and last image (they should be nearly identical as the last image is taken after a 360^o rotation of the specimen) to make sure the specimen did not move or the scan settings did not change during data acquisition.
3. Calculate the center of rotation (COR) by selecting the "Center of rotation" tab, and clicking "Start"; use options "Automatic" COR finding with "High Quality" accuracy, and "Dual" (upper and lower) slice selection for COR calculation.
4. Select the sample volume to be reconstructed: select the "Volume" tab, and edit the volume selection windows using the thumbnails.
5. Perform reconstruction to create the volume file containing 3D data by clicking "Start".

3. Data Processing / Segmentation

Note: This section describes the steps to be taken to prepare the reconstructed data for further processing in the program RooTrak to track roots as they branch out through soil, and isolate the roots from any surrounding material to produce a stack of binary images of just the root itself.

1. Processing of volume data in ImageJ to prepare a RooTrak processable image stack:
   1. Load the volume file into ImageJ.
   2. Optimize image contrast between root and soil by adjusting brightness and contrast settings (Click Image/Adjust/Brightness/Contrast). When the region of interest within the image is visible and clearly distinguishable, the settings are considered optimized.
   3. Save as an image stack in jpeg, bmp, or png format.
2. Processing in RooTrak to segment the root:
   1. Load image stack into RooTrak (go to "Tools" tab, and press "Tracker").
   2. Set seed points inside root: click several points inside each of the pertinent root sections visible in the top view slice of the volume data.
   3. Set tracker parameters "Smoothness" and "Similarity" to 0.3 and 0.8, respectively.
   4. Run the tracking function. This will follow the root from the top image slice all the way to the bottom slice.
   5. After viewing the volume data, select the number of slices according to the useable data volume; in this case, tracking was stopped at 200 slices, equivalent to a depth of 6.2 mm, where root boundaries became ill-defined (the image of the tracked root started to blend into that of the soil).
      Note: The image stack produced will be saved automatically to wherever the output directory was created.

4. Volume and Surface Analysis

Note: This step describes the isosurface mesh generation from the image stack created by RooTrak.

1. Convert image stack from RooTrak into a binary image format in ImageJ. Select "Process", then "Binary", then "Make Binary".
2. Use the open-source ImageJ plugin, BoneJ, to create the triangular mesh; in ImageJ select "Plugins", then "BoneJ", then "Isosurface".
3. Set "Resampling" and "Threshold" to 6 and 120, respectively (Default settings). Check "Show surface", and press the "OK" button.
4. On the "3D viewer" click on the File tab, then on "Export surfaces", then save as "STL (binary)".
5. Open imeshJ, select the STL file and enter voxel size in microns. Click "Calculate Surface Area" to acquire total sample root surface area in mm². Similarly, click "Calculate Volume" to obtain total sample root volume in mm³.

Results

The specimen consisting of two stems of the native grass Prairie dropseed (Sporobolus heterolepis) and the original soil around it was taken from a residential area and placed in a small tube-shaped holder seen in Figure 1. The reconstructed data voxel size was approximately 31 µm x 31 µm x 31 µm. The reconstructed volume file was used to create a stack of images from a selected orientation (top view) using the open-source image-processing program ImageJ 1.6^...

Discussion

A combination of X-ray computed tomography and several open-source programs proved to be a powerful combination to noninvasively image plant root samples, segment root data, and extract quantitative information (surface area and volume) from the 3D data. Our ability to visualize and measure features is always limited by scan resolution, as well as by limitations of the RooTrak software. However, scan resolution was sufficient to capture the majority of the features of the sample in this study, and RooTrak was able to suc...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
X-Tek/Metris XTH 320/225 kV	Nikon Metrology		X-ray tomography scanner
Inspect X	Nikon Metrology		Instrument control software
CT Pro 3D	Nikon Metrology		Reconstruction software, version XT 2.2
VG Studio MAX	Visual Graphics GmbH		Visualization software for 3D volumes, version 2.1.5
ImageJ	Open-source		Image processing and analysis software, version 1.6
RooTrak	Open-source		Root segmentation software, version 0.3.1-b1 beta
imeshJ	EMSL		MATLAB script developed by the authors
Prairie dropseed grass sample			Sample obtained from ground in residential area