A High Resolution (<1 micron) Computer Controlled Imaging SystemRichard W. Zobel

USDA-ARS Appalachian Farming Systems Research CenterRhizoecology - 1932-12000-004-00D1224 Airport Rd Beaver, WV 25813

Introduction/Abstract One difficulty in the routine analysis of fine root diameter response to environmental stimuli is the lack of scanners with resolutions above 189 p mm-1 (4800 dpi). A 10% change in diameter of a 0.1 mm diameter root is 0.01 mm or 2 pixels at 189 p mm-1. Because of innate variability, images need to have a resolution that has a 4 pixel shift when a 10% change occurs (i.e. 378 p mm-1 or better). Some pasture grasses have fine roots in the 30 micron diameter range, requiring resolutions of 1324 p mm-1 or a relative pixel size of 0.8 micron. We have achieved a resolution of 1600 p mm-1 by coupling a computer controlled stepper motor X-Y-Z system to a Zeiss dissecting microscope and a 14 Mp digital camera. The system can be programmed to do image slicing, thus providing high resolution images of three-dimensional objects without depth of field artifacts.

X -Stepper Motor

Moveable Stages

Light Box

Z-Stage

Camera

Dissecting Microscope

Limit Switch

Root holding tray

Materials:•Dell computer running EMC2 (linuxCNC.org) in an Ubuntu 8.04 environment.•Probotix three axis stepper motor kit

(http://www.probotix.com/3_axis_stepper_motor_driver_kits/)•Microscope: Zeiss, Stemi 1200 with zoom settings of 0.65 (204 p mm-1) to 5.0 (1601 p mm-

1). Resolutions are those with a Kodak DCS ProSLR/n (13.5 MP) camera body .•WinRhizo v 2007 (RegentInstruments.com) was used to determine root length and diameter class.•Non linear regression, Extreme Value or Triple Gaussian models (TableCurve 2D from Systat).

Figure 1a. Image G3FN4425, image resolution is 204 p mm-1. This is prior to thresholding and smoothing. Note the presence of root hairs.

1a. 1b.

Figure 1b. After thresholding (128) and median smoothing (5 pixel radius) of image G3FN4425. Thresholding and smoothing have eliminated most of the root hairs.

Results:As equipped, the XYZ imager is able to take images at resolutions from 204 p mm-1 to 1601 p mm-1 (5,186 dpi to 40,325 dpi). This gives image relative pixel sizes of 4.9 to 0.6 microns. The requirement of a 0.8 micron pixel size can be obtained at the 4.0 zoom setting of the microscope (0.77 microns). A digital camera with a smaller pixel density will give lower resolutions. The stepper motors, as set up, average 0.5 microns per step. The relatively crude x-y table has a 2 mm variance, partially negating the accuracy of the stepper motors. The program for imaging a whole root system moves the tray from the 0, 0, 0 position (z-direction) -23.9 mm (camera up), y-direction 214 mm (tray toward the observer), and x- direction 208 mm (tray to the left). The system takes an image and then moves -22.4 mm X and takes another image. After nine images the tray is moved -14.71 mm y and the sequence reversed. This is done 13 times for 117 images and a cumulative area of 90.05 % of the tray surface (less than 20 minutes). Assuming uniform distribution of the roots, this means that a correction factor of 1.11 times measured root length gives actual root length.

WinRhizo is run on each image after adjusting the threshold and applying median smoothing. When the resulting 117 data records are combined, the new record is nearly identical to that from a full scan image of the roots. The resulting record yields a typical, length by diameter class distribution profile (Fig. 2). The non-linear regression model “Extreme Value” consistently gives the best R values for data from WR analyses of whole root systems. If the length distributions of individual images are plotted, a series of discrete peaks are observed (Fig. 3). From another study, it is known that the shoot-borne roots are from 0.2 to 0.5 mm in diameter, lateral roots are from 0.1 to 0.15 mm in diameter, and the secondary laterals are from 0.05 to 0.09 mm in diameter. The peaks in figure 3 have their maximum values in the same ranges.

When WR data from a single image (G3FN4425.tif) is plotted separately (Fig. 4) the relationships between the peaks is more clear. The shape of the peaks suggests two overlapping and one isolated normal curves. If three simultaneous Gaussian curves are fitted to the data, a very close fit is found (Fig. 4, Table 1). A Gaussian curve is the normal curve of statistics and the height, mean diameter and standard deviation can be calculated (Table 1). This information can then be used to accurately model (reconstruct) the root distributions in the future. Only very complex images with many roots can not be modeled with one or more simultaneous Gaussian curves. As the overlapping of the curves of figure 3 suggest, more complex images tend toward the extreme value model of figure 2.

Methods:Tiller derived plants of five perennial ryegrass clones were grown in soil for 1 month, roots washed out, cut into approx. 2 cm lengths and stained with 0.1 % toluidine blue. The roots and 400 ml tap water were placed in a 200 x 200 mm square dish spread out and automatically imaged in 117 separate images with no overlap. A wire (30 mm long and 0.201 mm diameter) was placed in the tray to provide a known reference for thresholding and smoothing. Thresholding at 128 and a 5 pixel radius median smoothing were applied sequentially to each image with Photoshop (Adobe.com), see Fig.1.

Figure 2. The length by diameter distribution plot for the combined data from images G3FN4374 to 4490. The plant was a clone of LpAc 40, the XYZ was set at its lowest resolution (204 p mm-1).

Figure 3. length by diameter distribution plots of four images selected from the composite of figure 2.

Figure 4. the length by diameter class distribution plot for image G3FN4425, and the triple Gaussian model for that data.

R = 0.97 Height Diameter Std. Dev

Curve 1 0.899 0.079 0.014

Curve 2 0.481 0.129 0.019

Curve 3 0.148 0.340 0.019

Table 1. Calculated coefficients, and model R value for the three modeled curves of figure 4.

Conclusion: The XYZ imager is capable of imaging a whole root system at higher resolutions than possible with scanners. By breaking the root system down and imaging smaller pieces, it is apparent that there are characteristic modelable parameters to the root system of perennial ryegrass and that roots fit into discrete classes. The diameters of different roots are normally distributed around a mean diameter, therefore: Root Systems Are Not Made Up of a Continuous Distribution of Diameters.

DiscussionThe routine calibrations of the XYZ imager, demonstrate that it is capable of pixel resolutions down to 0.6 microns (1601 p mm-1 in images), which should be suitable for many fungal hyphae and small soil flora and fauna. The 0.5 micron per step movement of the camera and the X & Y stages should allow image slicing and multiple imaging to follow root growth or the movement of small insects.

In previous work, it was noted that separating root systems into major branches and imaging separately, reduced the apparent complexity of the root system. With a scanner based system, this is time consuming and restricted to relatively low resolutions. The diameters calculated with WR are similar to those obtained at 100x and visual measurement. Use of this system with intact well spread out root systems should allow detailed topological studies.

All the curves in figure 3 appear to be suitable for Gaussian curve modeling. As in corn, younger shoot-borne roots are thicker than older (initiated earlier) ones. In perennial ryegrass, dominant lateral roots have diameters in the range between 0.2 mm and half the diameter of their parent shoot-borne root, while secondary laterals obey the half the parent diameter rule. True laterals maintain their clone specific diameters (0.122 mm in clone LpAc 40).

The basic underlying root length by diameter distribution is that of a normal curve. When a root branches, the distribution is that of a double normal curve. Given a limiting minimum diameter of 0.06 mm and multiple shoot-borne roots, an extreme value curve for the total distribution would be expected. This refutes the theory of a continuous distribution of root diameters.