Fig 1.
Labelled image of a wheat root cross section.
The epidermis is the outermost single layer of cells of the root. The cortex is the second outermost region and houses the largest cells in the root. The endodermis is the single layer of cells forming the boundary between the cortex and the stele. The stele is the set of cells inside the endodermis. Finally, the metaxylem, denoted in the figure by M.X, is located in the centre of the root and is repsonsible for water and nutrient transport.
Table 1.
Primary statistical features captured by the RootAnalyzer algorithm.
A selection of these is discussed in detail in the Output section.
Fig 2.
Flowchart outlining the steps of the algorithm.
First we acquire the images using a Leica AS LMD laser dissection microscope with a DFC 480 camera and then segment them using a local thresholding technique. Cells are detected using a distance transform of the segmented image and are then initially classified by their area. A number of subsequent steps then correctly classify cells into the four main regions; cortex, stele, endodermis and central metaxylem.
Fig 3.
Visual depiction of the local thresholding approach.
If the intensity at a pixel (x, y) differs from the mean intensity inside the window by more than some pre-defined threshold, we conclude that it is a foreground pixel.
Fig 4.
Challenging images make local thresholding a necessity.
(a): Input images with regions of varying intensity. (b): Segmentation result using RootScan’s global thresholding method. (c): Segmentation result using RootAnalyzer’s local thresholding method.
Fig 5.
(a) Binary segmented image. (b) Distance transform applied to the segmented image. Peaks correspond to the centers of cells. (c) Cells are detected by interpolating around each location where the binary image changes from background to foreground.
Fig 6.
Finding the central metaxylem.
Left: Collection of all small cells. Middle: Histogram showing the distance from the centre of each cell to the centre of the root vs frequency. The peak at low distances corresponds to cells inside the stele and endodermis. Right: Keeping cells whose distance is greater than the illustrated threshold provides a first estimate of stele and endodermis cells.
Fig 7.
Sample result of RootAnalyzer applied to a maize image.
(a) The original maize image. (b) Illustration of the classification process. Deep blue represents a higher likelihood of belonging to the stele or endodermis. (c) The result after all steps of the algorithm are complete. Blue cells belong to the cortex, magenta cells are aerenchyma, red cells belong to the stele and endodermis, cyan cells are protoxylem elements and yellow cells are central metaxylem.
Fig 8.
Sample of the cheesewheel approach.
(a) The entire root section with partitioned cortex. (b) Cheesewheel partition of the stele and endodermis. In (a) the inner boundary of the inner-most annulus is created using the boundary of the endodermis as a guide. Similarly, the outer boundary of the outermost annular region is created using the boundary of the epidermis as a guide.
Fig 9.
RootScan and RootAnalyzer applied to a sample maize image.
(a) The original image. (b) Results from RootScan. (c) Results of RootAnalyzer.
Fig 10.
Sample results from the RootAnalyzer algorithm.
(a) the entire root section analysed. The magenta rectangle in the bottom image denotes a small piece of segregated root that the algorithm has failed to capture. (b) zoomed versions of the respective stele regions highlighting cortical cells (red), epidermal cells (orange), stele cells (blue) and central metaxylems (green).
Table 2.
Primary statistics captured by the RootAnalyzer algorithm when applied to the example maize image.
The error of each measurement, as compared to the manually obtained ground truth, is also included.
Table 3.
Primary statistical averages and errors thereof of features captured by the RootAnalyzer algorithm; averages are over all 15 wheat images (columns 3–5).
In columns 6 and 7, the degrees of error result from using a subset of images only: 13 high quality images and 2 poor quality images, respectively.
Fig 11.
Two images from the set of 15 showing damaged roots.
The two images depict damaged root cross-sections with missing or displaced epidermis and cortex regions. The damaged sections are major factors contributing to the relatively large errors in cortex (and presumably epidermis) cell calculations.
Fig 12.
Percentage errors in evaluating average area of cortex cells.
13 of the 15 images produce a much lower degree of error in average cortex cell area calculation. The two red markers of significant error correspond to poor quality (i.e., damaged) root images.