Conditions for X-ray CT scanning
Assuming the application for large-scale root phenotyping of crops, a scanning time of 10 min for each plant sample was determined as the X-ray CT condition. Because additional time was required for machine operation, start-up of X-ray generator, and saving the CT images, the actual time was 15 min for a single sample. Therefore, 32 samples could be processed in a day, provided a working time of 8 h a day. For example, weekly scanning could process 160 individuals per week, which is sufficient to perform large-scale phenotyping. To observe the RSA development of rice continuously before the roots reached the pot wall, the pot diameter and depth were set as 20 cm and 25 cm, respectively, based on the maximum size that can be scanned by the detector of the CT scanner used in this study. The scanning conditions were as follows: each scan digitally obtained 1200 projections using a signal averaging of two frames over 360° without binning (pixel detector resolution: 3000 × 3000) at 4.0 frames per second (fps). Finally, 860 horizontal slices of pixel resolution 1024 × 1024 were computed. The final spatial resolution was 300 μm, corresponding to a total volume of 30.72 × 30.72 × 25.8 cm3. These conditions can simultaneously process the scanning and the reconstruction steps in the CT system we used, and are thus advantageous for a rapid CT scan.
To obtain images containing a clear root shape, we determined the soil substrate for CT scanning. We used the upland rice cultivar Kinandang Patong (KP) as the test sample to satisfy the requirement of CT scanning. Moreover, KP was expected to be relatively easy in isolating the root segments from CT images because upland rice usually has thicker roots than lowland rice [31]. To select the suitable soil substrate for CT scanning of rice roots, we examined the CT images of the roots of KP grown in five different soil substrates, namely, calcined clay, volcanic ash soil, andosol, alluvial soil, and sand (Additional file 1). Among these five types, calcined clay exhibited the clearest root shape in the CT images (Additional file 2). Based on this result, we decided to use calcined clay for CT scanning of rice roots.
Because the tube voltage and the current in X-ray CT scanning affect the CT image quality, we determined them such that they exhibit the highest root-to-soil contrast. Calcined clay was packed into the pot and saturated with a hydroponic solution based on Kimura B solution, which is used for rice cultivation [32]; nitrate and ammonium concentrations were adjusted for upland condition. Further, KP was grown in the pot for five weeks in a growth chamber, and was subjected to CT scanning. The representative CT images are shown in Fig. 1. The inside of the pot was invisible in the 3-D reconstructed volumes (Fig. 1a). In the horizontal and the vertical slices (Fig. 1b and 1c, respectively), the roots are visible as dark pixels. The pixels of lower values were colored in black, which indicates lower X-ray absorbance in the rice roots compared with that of calcined clay. To evaluate the influence of tube voltage and current on the CT image quality, we scanned the pot with the tube voltages of 125 kV, 150 kV, 175 kV, 200 kV, and 225 kV and tube currents of 100 μA, 200 μA, 300 μA, 400 μA, and 500 μA. The scaled-up images of all these combinations are shown in Fig. 2. Apparently, higher voltage and current produced the images highest contrast. This was supported by the fact that the peak signal to noise ratio (PSNR) increased with higher tube voltage and current, attaining the highest value at 225 kV and 500 μA. These results conclusively indicated that the voltage of 225 kV and the current of 500 μA were the best combination for rice root scanning under our conditions.
Image processing
To visualize the rice root segments automatically, we developed the image processing pipeline. The developed pipeline involves the following two steps: (1) a 3-D median filter process to increase the root-to-soil contrast and (2) an edge detection process to dilute the soil-like texture but retain the root-like structure.
The first step is to increase the root-to-soil contrast to reduce noise in the CT images. Noise is caused by mineral particles, void in the soil, and short scan time because we selected a short scan time in CT imaging for high-throughput imaging. To reduce the noise level, we applied a 3-D median filter to the CT images. Fig. 3a shows vertical slices with five different kernel sizes of 1, 3, 5, 7, and 9. An image processed with the kernel size of 1 is equivalent to a non-filtered image. We calculated the PSNR of each condition and found that the PSNR was the highest for the 3-D median filter of kernel size 7. The image with the kernel size of 9 was the most blurred. Thus, we determined the kernel size as 7 for this process based on these results.
The images were processed by the following steps to segment the root area in the images. First, the value intensity of the CT images was inverted because the CT images had soil voxels with higher value intensity and root voxels with lower value intensity. Next, we subtracted the blurred slices from their corresponding non-blurred counterparts to zero soil value intensity. The pixels, whose image brightness changes sharply would be isolated as root segments, which is a simple algorism of edge detection. Fig. 3b shows vertical slices subtracted with blurred images with various kernel sizes. Because an image processed with the kernel size of 1 is the same as a non-filtered image, it results in all-zero images. In the image processed with a kernel size higher than 5 (Fig. 3b), signals were observed at the positions where the root is located in Fig. 3a. The areas of root became larger with the increase in the kernel size (Fig. 3b). Furthermore, the area of signals in the image with the kernel size of 21 was the same as the area of root in Fig. 3a with the kernel size of 1. Thus, we decided to use the kernel size of 21 based on these results.
After cropping the region of the inside of the CT images to eliminate the effects of pot wall on RSA development, all the processed slices were stacked to construct 3-D volumes. The horizontal projection and 3-D animation of the 3-D rendered volume are shown in Fig. 3c and Additional file 3, respectively. Furthermore, the RSA in the soil was successfully visualized; however, non-root segments were additionally recognized in the image and the movie. Because the root segmentation depends on contrast difference, all the voids in the soil are visualized. The non-root segments at the bottom were caused by collapsing of soil, and the segments at the top were cracks caused by plant growth and incompletely packed soil close to the ground surface. Small particles appearing everywhere were voids or water gradient in the soil. To remove small non-root segments, we used a thresholding and size opening method. Thresholding cut off connections of each segment, and the small segments were removed by a size opening filter. Filtered horizontal projection and 3-D animation are shown in Fig. 3d and Additional file 4, respectively. Root segments were unaffected by these processes in this case, but the risk of erasing of small root segments by thresholding and size opening existed. For this reason, we did not use the thresholding and size opening filter in this study.
We implemented the algorism by python script (Additional file 5) and measured the processing time with different hardware (Table 1). All the hardware we tested took less than 8 min for image processing. The processing time depended on the central processing unit, and the fastest processing time of 2 min was achieved using Intel® Xeon® E5-2650 v4. Because python is an interpreter language, batch operation is easily executable.
Scanning time and metal filters to reduce X-ray doses
X-rays affect the plant growth [22], which is a problem for 4-D root phenotyping with X-ray CT systems, because repeated CT scanning increases the cumulative X-ray dose. To reduce the X-ray doses, we investigated the influence of scanning time and metal filters on the CT image quality.
The scanning time is determined by the projection number, signal averaging number, exposing time, and binning size. Under our conditions, the projection number, signal averaging number, and binning size can be changed to reduce the scanning time. In a scanning time of 10 min, each scan digitally obtains 1200 projections using a signal averaging of two frames over 360° without binning (pixel detector resolution: 3000 × 3000) at 4.0 fps. At the fastest scanning, each scan digitally obtains 600 projections using no signal averaging over 360° with 3 × 3 binning (pixel detector resolution: 1000 × 1000) at 18.0 fps. In this condition, scanning is performed in 33 s. We scanned a pot in which KP was cultivated for eight weeks under eight scanning conditions, and obtained 600 or 1200 projections using no signal averaging or a signal averaging of two frames over 360° with or without binning. The horizontal projections of the processed CT images are shown in Fig. 4. We observed similar RSA in all conditions, despite the degraded image quality at faster scanning conditions. Based on the result, the decision to shorten the CT scanning time for low X-ray doses was rejected.
Another method to reduce the X-ray doses is to use metal filters. The X-ray beams have a range of wavelengths. Because X-rays of longer wavelength have lower energy, their penetration ability is low, and are thus absorbed at the material surface, resulting in high X-ray dosage. Generally, metal filters are used to reduce the proportion of low-energy X-rays. To reduce the X-ray doses to plants, we evaluated the influence of copper (Cu) filters on the CT image quality. The scaled-up horizontal slices of unprocessed CT images without and with 0.5 mm, 1.0 mm, and 2.0 mm Cu filters are shown in Fig. 5. The noise level increased with the increase in the thickness of the filters. On the contrary, the quality of the image-processed horizontal projection using Cu filters were very similar to those with no filters; however, we observed a small increases in noise in the CT image when using the 2.0 mm Cu filter (Fig. 5). These results indicated that the 1.0 mm Cu filter is effective in reducing the X-ray dosage on plants under our conditions. Therefore, we used the 1.0 mm Cu filter in this study.
Influence of X-ray dose on rice growth
To evaluate the influence of X-ray CT exposure on rice growth, we estimated the X-ray doses using Rad Pro Dose Calculator (http://www.radprocalculator.com/). When a tube voltage of 225kV and current of 500 μA were applied to the material placed at 900 mm from the X-ray source using a 0.5 mm Cu filter, the X-ray dose of the material was estimated as 0.55 Gy/hr. At a scanning time of 10 min using an 1.0 mm Cu filter, the dose to rice plants was estimated as less than 0.09 Gy per scan. Because 0.09 Gy is sufficiently lower than 33 Gy, which is the threshold affecting plant growth [22, 33], it was considered that sequential X-ray scanning does not pose a problem for rice growth. In the case of rice, it was revealed that daily scanning with a dose of 1.4 Gy for nine days, i.e., total dose of 12.6 Gy, did not negatively impact the rice growth [22]. A simple arithmetic calculation indicated that 140 scanning procedures are permissible in our scanning condition, if 12.6 Gy is the upper limit for X-ray CT exposure.
To evaluate the influence of sequential X-ray doses on plant growth, KP was cultivated for two weeks and subjected to daily CT scanning for seven days. The results indicated no apparent differences between shoot and root shape of mock- and X-ray-treated plants at 21 DAS (Fig. 6a–b); also in addition, we quantified the shoot and the root traits. There was no significant difference in plant height, total root length, shoot dry weight, and root dry weight. (Fig. 6c–f). These results indicated that X-ray doses in our scanning condition do not constitute any problem for rice growth.
Four-dimensional visualization of root development
To evaluate the fully automatic visualization method in this study, we monitored the dynamics of root development of KP for three weeks. We cultivated KP for one week, and KP was subjected to daily CT scanning for three weeks. The processed horizontal projections and 3-D movie are provided in Additional files 6 and 7, respectively. From seven to 13 days after sowing (DAS), the root shape in the image was hazy but the daily root growth was observed. From 14 to 20 DAS, the root shape became bolder and the root length increased rapidly. Many root tips went outside the scope at 20 DAS. From 21 to 27 DAS, the root shape became increasingly bolder but the general shape of RSA remained unchanged. They indicated that our pipeline visualized the root development dynamics of KP for three weeks from seven DAS.
Verification of root fragments in the processed CT images
To verify the length of root diameter detectable with our pipeline, we implemented the wired basket method. The basket method is used to evaluate the rooting angle of rice cultivars by counting the proportion of roots penetrating the bottom of the basket [34, 35]. As a modified approach, we used the basket, whose inside was wired with nylon monofilament at intervals of 1.5 cm, to keep in situ RSA when the basket was unearthed from the ground and the soil was removed. More information on the procedure is available in the Methods section. We cultivated rice plants for 21 days and then unearthed the basket. Because we observed that 21-DAS KP has many roots of different thicknesses (Additional file 6), 21-DAS rice was suitable for verifying the length of the detectable root diameter. To exclude the influence of root distribution, we used three genotypes that had different RSAs, [36], namely, KP (thick and deep-root type), IR64 (thin and shallow-root type), and Dro1-NIL (thin and intermediate-root type). To visualize the roots in the soil, we scanned the rice plants using the X-ray CT scanner and performed image processing. The vertical projections are shown in the left column in Fig. 7a. After scanning, to obtain a visible root image keeping in situ RSA, we unearthed the baskets and removed the soil from the baskets by washing with tap water. The images shot from directly above the basket are shown in the middle column in Fig. 7a. We compared the CT and the camera images, and traced the crown and the radicle roots in the projection images (right column in Fig. 7a). There were 68 detectable (solid line) and 12 non-detectable (dash line) roots in the processed X-ray CT images, compared with the camera images. We collected the root segments, scanned them with a 2-D scanner, and measured the root diameters using ImageJ plug-in SmartRoot [37]. We compared the root diameters of detectable and non-detectable roots, and found that many roots with a diameter of less than 0.3 mm were not visualized by our method, irrespective of the RSA type (Fig. 7b). To visualize all radicle and crown roots, we used a smaller pot of diameter 16 cm, and adjusted the source-detector distance and source-rotation axis distance to be 800 mm and 407 mm, respectively. We performed the wired basket assay again, and found 82 detectable roots but did not find any non-detectable roots (Fig. 7c). The diameter of the detectable roots was higher than 0.2 mm (Fig. 7d). These results indicated that the detection limit of roots can be determined by adjusting the pot diameter, source-detector distance, and source-rotation axis distance. In this condition, the X-ray dose per scan was estimated to be 0.44 Gy. A simple arithmetic calculation indicated that 28 scanning procedure are permissible in our scanning condition, if 12.6 Gy is the upper limit for X-ray CT exposure.