Identifying biological landmarks using a novel cell measuring image analysis tool: Cell-o-Tape
© French et al; licensee BioMed Central Ltd. 2012
Received: 23 November 2011
Accepted: 2 March 2012
Published: 2 March 2012
The ability to quantify the geometry of plant organs at the cellular scale can provide novel insights into their structural organization. Hitherto manual methods of measurement provide only very low throughput and subjective solutions, and often quantitative measurements are neglected in favour of a simple cell count.
We present a tool to count and measure individual neighbouring cells along a defined file in confocal laser scanning microscope images. The tool allows the user to extract this generic information in a flexible and intuitive manner, and builds on the raw data to detect a significant change in cell length along the file. This facility can be used, for example, to provide an estimate of the position of transition into the elongation zone of an Arabidopsis root, traditionally a location sensitive to the subjectivity of the experimenter.
Cell-o-tape is shown to locate cell walls with a high degree of accuracy and estimate the location of the transition feature point in good agreement with human experts. The tool is an open source ImageJ/Fiji macro and is available online.
KeywordsImage analysis Confocal Arabidopsis Software Quantification
Software capable of aiding the manual phenotyping of plants has become the subject of much interest in recent times . There has been a push to develop high throughput imaging platforms [2–5] in an attempt to make use of digital-image-derived data for phenotyping, which has lagged behind the vast quantities of genomic data now available . To cope with the resulting surge in raw data, new tools are being developed to automate image analysis processes and assist experts in performing traditionally labour-intensive tasks such as cell scale segmentation [7, 8], root length measurement  and architecture description [9, 10]. Additionally, software tools are opening up new ways to analyse data which is not practical manually, using techniques such as optic flow  to examine growth at the pixel-scale, and 3D segmentation of confocal time series datasets .
The tool adopts a semi-automated approach, which requires the user to define a region of interest by specifying a profile line through the cells. The user can improve the quality of the output if necessary by taking part in an iterative cycle of output refinement, adjusting parameters as necessary and even manually enhancing particularly noisy image features, although this is largely unnecessary. This semi-automatic approach, when compared to fully automatic approaches , has the advantage of allowing the user to define exactly where on the image the tool operates, and allows the user to improve the quality of the output using their expert knowledge, whilst performing the main labour-intensive processing in an automated, objective way.
Here, we describe the background to the problem and the algorithms used, then evaluate the success of the tool by considering a common task in root biology; identification of the transition point into the elongation zone. Cell-o-Tape's abilities to locate walls and locate this feature point are compared to results obtained by human experts working with confocal images of roots of Arabidopsis thaliana.
The exemplar application in this paper is the study of plant root growth, a topic which has seen increased activity as a result of the recent surge of interest in both global food security and bio-energy production. It is complex and expensive to measure plant roots at the cellular scale over time with acceptable biological replication using traditional methods, hence the requirement for automated tools.
The growth of the plant root is driven by distinct processes of cell division and cell elongation in separate developmental zones of the organ . In the root meristem, cells differentiate and divide from precursor stem cells over a period of ~100 hrs. Cells then proceed through a transition zone (TSZ) before starting a process of rapid elongation in the elongation zone wherein they increase in size around 15 times over 6-8 hrs [15, 16]. As root growth is driven by both the number of cells produced by cell division and the extent of their elongation, the identification of both cell size and cell number is crucial to characterisation. Existing growth analysis software typically defines kinematics of growth using pixel-scale [11, 17] or cell-scale motion tracking [8, 18] over image sequences, but the Cell-O-Tape approach requires only an individual image. Due to the constrained, linear growth process, a single cell-resolution image of the root contains all the growth stages and cellular scale information required, which can be thought of as a 'timeline snapshot' of these cell-scale developmental processes. Therefore, extracting cell counts and sizes from these images can provide valuable information about the underlying processes involved.
Despite the linearity of the root tissue, cell sizes in the meristem and elongation zone are notoriously heterogeneous and the measurement of a large number of cells is required to accurately characterise the nature of any growth phenotype, hence a high labour cost in performing the required measurements. Automating this process should therefore make the process more efficient and reproducible.
In addition to general cell measurements, there is one phenotypic feature of particular interest when examining cell elongation in Arabidopsis roots. The switch-point from the division zone to the elongation zone is a key observation of many growth phenotypes, usually identified as the point in the TSZ where cells start expanding rapidly, and hypothesised as the point marking the end of mitotic activity. However, there is a subjective judgement present when this point is identified by experts, as it is usually determined simply by a visual inspection of the image. In this work, as an extension to the cell-measuring capabilities of the tool and to demonstrate the accuracy and value of the cell-size data produced, we hypothesise that this point can be determined automatically and objectively from the underlying cell positions and lengths. A higher-throughput bias-free mechanism to determine this point would significantly aid growth studies. Additionally, combining the raw data and/or TSZ point calculated by Cell-o-Tape with other complimentary software kinematic approaches (eg. ) would provide a powerful suite of tools with which to analyse root growth.
Usage of the tool proceeds as follows. The approach requires a manually defined line through a cell file or group of cells, allowing the user to define which file of cells the software will measure. Once this linear region is defined using Fiji's "segmented line" tool, the macro fits a spline, producing a smooth line through the defined points. This profile line can pass through any file of cells (so long as cell walls are sufficiently clear) and be of any length; note though that to use the TSZ estimating feature which is described later, the profile line must extend sufficiently far into the elongation zone such that the change in cell length can be detected. To use this feature, ensure at least two substantially elongated cells (~4x meristem cell length) are included.
Each point along the fitted line is tested for its viability as a cell wall location. The current pixel under consideration shall be referred to as p t
A window of pixels of a fixed size is examined, centered on p t .
The minimum and maximum intensity levels (i min , i max ) in the window are stored, ignoring p t
If the intensity at p t ≥ i max and the intensity at p t ≥ i min * stringency, then mark p t as a cell wall location.
Hypothesised wall locations are marked on the image and presented to the user. This process of user interaction followed by calculation and presentation of the results occurs in a loop so the user is able to visually refine the results until all the cell walls are detected. Empirical use of the software has shown that excellent results can be achieved this way in a minimal time on typical confocal images. See Additional file 2: Video 1 for an example of the tool in use.
The user is also presented with tools for manually enhancing walls in the image, and suppressing noise, if the data is of such low quality or contrast that they cannot all be picked out automatically. The tool is then re-run and the new walls detected.
The size transition point is calculated automatically, presented on a graph and marked on the output image (Figure 3(a)); additionally the distance from the user-specified start point (in our case the cortex and endodermis initial) and cell count to this point are reported in the log window, along with all the individual cell length measures along the line (e.g. Figure 2). These heterogeneous output visualisations provide the user with multiple error checking opportunities, and additional evidence for any claims they make in publications. By saving the profile line region-of-interest from Fiji and noting the parameters for Cell-o-Tape, reviewers can re-run the algorithm on raw image data to verify claims made by authors, if necessary.
Results and discussion
First, the accuracy of the cell length measures generated by the tool was compared to a manual approach. The detected lengths, as derived from the location of cell walls, were compared to manual measures of length. A root-mean-square error of 0.3 μm (n = 66 cells, length range ~10-90 μm) shows a good agreement with groundtruth manual length measures.
To assess how sensitive the tool is to how far the profile line extends along the root when calculating the TSZ point, two variations were considered. In the first case, a situation where the user does not extend the profile line far enough into the elongation zone was considered. To test this, the TSZ point was recalculated for the 26 images, but discarding the final two measurements along each profile line (typically the two largest cells); see Additional file 5. On 18 occasions (70%) this made no difference to the TSZ point estimation. On the 8 images where the location of the point differed, the average error in cell location number was 7.5 cells (15%). So, while on most occasions no extra error is introduced if the profile line is terminated too early, there is a chance of a significant error. On examining the cell ID versus cell length graphs, this error is likely being caused by the graphs becoming overly compressed on the vertical axis, caused by the omission of the longer cells, therefore invalidating the assumption of a detectable corner point  as well as making the technique more sensitive to noise. Therefore, it is recommended that the profile line be extended to include at least two substantially elongated cells (~4x meristem cell length). This would seem reasonable, as any approach that estimates a feature such as the TSZ point would need a sufficiently extensive data set as a starting point.
The complementary situation, where the profile line extends further towards the mature zone, was also examined. This was achieved by defining the profile line further towards the mature zone on a composite image (Additional file 6). It was expected that this case would not cause problems for the calculation of the TSZ point as there should still be a detectable corner point on the graph . As can be seen in Additional file 6, the TSZ is still located correctly on this longer profile line, and the more elongated cells still correctly delineated.
We have presented Cell-o-Tape, a tool primarily designed for automatically counting and measuring cells along a defined profile in confocal images. Cell-o-Tape can also identify a significant change in cell length along the profile, a facility which can be used to estimate the location of the start of rapid cell elongation along a file of cells in the Arabidopsis root, as shown in our example application.
The tool was found to be able to accurately measure cells compared to a human expert. A method was presented to determine the location of a TSZ feature point by detecting a 'break point' on a cell number versus cell size graph. This process was found to estimate the position of this TSZ landmark point in agreement with manual expert measures as well as the experts agree with each other. In addition, reporting the location of the TSZ in terms of distance along the root yielded a higher Williams Index than when the number of cells was used as the comparison metric, although this may be due to the higher variability between the experts in this case (Figure 4). Additionally, it was observed that the experts sometimes agreed on the TSZ landmark location using distance, but provided a differing cell count. This highlights the subjective nature of counting cells, both in terms of simple miscounting, and misinterpreting noise or 'ghosting' from surrounding cell layers as erroneous cell walls; these factors are not present when reporting the location in terms of distance along the root. Thus reporting the TSZ location using distance would seem to be a more reliable measure in general than providing a cell count, as it is not as sensitive to missed cells. In both cases, the tool is seen to agree with the experts, and thus provides an objective and reproducible way to perform these measurements.
We recommend using the tool as an efficient way to measure cell lengths and count cells along cell files with minimum interaction required by the user, and also to provide an impartial estimate for the TSZ feature location in Arabidopsis roots, alongside the user's expert but subjective opinion. The graph the tool presents can also be used to support this location, and this representation also helps alert the user to artefacts in the process, meaning they can then re-examine the image if necessary. Although developed for Arabidopsis root images, we see no reason why the tool could not be used on other confocal images with similar content.
Availability and requirements
Project name: Cell-o-Tape
Project home page: http://www.cpib.ac.uk/software/
Operating system(s): Any Imagej/Fiji supports
Programming language: ImageJ/Fiji macro language
Other requirements: Developed on Fiji 1.45r
We would like to acknowledge Dr Susannah Lydon for the name of the tool.
The authors acknowledge EPSRC and BBSRC CISB programme funding to CPIB.
- de Souza N: High-throughput phenotyping. Nat Meth. 2010, 7: 36-View ArticleGoogle Scholar
- French A, Ubeda-Tomás S, Holman TJ, Bennett MJ, Pridmore T: High-throughput quantification of root growth using a novel image-analysis tool. Plant Physiol. 2009, 150: 1784-1795. 10.1104/pp.109.140558.PubMed CentralView ArticlePubMedGoogle Scholar
- Durham Brooks TL, Miller ND, Spalding EP: Plasticity of Arabidopsis Root Gravitropism throughout a Multidimensional Condition Space Quantified by Automated Image Analysis. Plant Physiol. 2010, 152: 206-216. 10.1104/pp.109.145292.PubMed CentralView ArticleGoogle Scholar
- Miller ND, Durham Brooks TL, Assadi AH, Spalding EP: Detection of a Gravitropism Phenotype in glutamate receptor-like 3.3 Mutants of Arabidopsis thaliana Using Machine Vision and Computation. Genetics. 2010, 186: 585-593. 10.1534/genetics.110.118711.PubMed CentralView ArticlePubMedGoogle Scholar
- Nagel KA, Kastenholz B, Jahnke S, van Dusschoten D, Aach T, Mühlich M, Truhn D, Scharr H, Terjung S, Walter A, Schurr U: Temperature responses of roots: impact on growth, root system architecture and implications for phenotyping. Funct Plant Biol. 2009, 36: 947-959. 10.1071/FP09184.View ArticleGoogle Scholar
- Houle D, Govindaraju DR, Omholt S: Phenomics: the next challenge. Nat Rev Genet. 2010, 11: 855-866. 10.1038/nrg2897.View ArticlePubMedGoogle Scholar
- Vylder J, Rooms F, Philips W: Segmentation of Cell Nuclei in Arabidopsis Thaliana Roots. Image Analysis and Recognition. Edited by: Campilho A, Kamel M. 2010, Berlin, Heidelberg: Springer Berlin Heidelberg, 207-216. 6112View ArticleGoogle Scholar
- Sethuraman V, French A, Wells D, Kenobi K, Pridmore T: Tissue-level segmentation and tracking of cells in growing plant roots. Machine Vision and Applications. 2011Google Scholar
- Lobet G, Pagès L, Draye X: A Novel Image Analysis Toolbox Enabling Quantitative Analysis of Root System Architecture. Plant Physiology. 2011, 157 (1): 29-39. 10.1104/pp.111.179895.PubMed CentralView ArticlePubMedGoogle Scholar
- Armengaud P: EZ-Rhizo software. Plant Signal Behav. 2009, 4: 139-141. 10.4161/psb.4.2.7763.PubMed CentralView ArticlePubMedGoogle Scholar
- van der Weele CM, Jiang HS, Palaniappan KK, Ivanov VB, Palaniappan K, Baskin TI: A New Algorithm for Computational Image Analysis of Deformable Motion at High Spatial and Temporal Resolution Applied to Root Growth. Roughly Uniform Elongation in the Meristem and Also, after an Abrupt Acceleration, in the Elongation Zone. Plant Physiol. 2003, 132: 1138-1148. 10.1104/pp.103.021345.PubMed CentralView ArticlePubMedGoogle Scholar
- Fernandez R, Das P, Mirabet V, Moscardi E, Traas J, Verdeil J-L, Malandain G, Godin C: Imaging plant growth in 4D: robust tissue reconstruction and lineaging at cell resolution. Nat Meth. 2010, 7: 547-553. 10.1038/nmeth.1472.View ArticleGoogle Scholar
- Quelhas P, Nieuwland J, Dewitte W, Mendonça AM, Murray J, Campilho A: Arabidopsis Thaliana Automatic Cell File Detection and Cell Length Estimation. Image Analysis and Recognition. Edited by: Kamel M, Campilho A. 2011, Berlin, Heidelberg: Springer Berlin Heidelberg, 1-11. 6754View ArticleGoogle Scholar
- Verbelen J-P, De Cnodder T, Le J, Vissenberg K, Baluška F: The Root Apex of Arabidopsis thaliana Consists of Four Distinct Zones of Growth Activities. Plant Signal Behav. 2006, 1: 296-304. 10.4161/psb.1.6.3511.PubMed CentralView ArticlePubMedGoogle Scholar
- Beemster GTS, Baskin TI: Analysis of Cell Division and Elongation Underlying the Developmental Acceleration of Root Growth in Arabidopsis thaliana. Plant Physiol. 1998, 116: 1515-1526. 10.1104/pp.116.4.1515.PubMed CentralView ArticlePubMedGoogle Scholar
- Beemster GTS, Baskin TI: STUNTED PLANT 1 Mediates Effects of Cytokinin, But Not of Auxin, on Cell Division and Expansion in the Root of Arabidopsis. Plant Physiol. 2000, 124: 1718-1727. 10.1104/pp.124.4.1718.PubMed CentralView ArticlePubMedGoogle Scholar
- Wuyts N, Bengough AG, Roberts TJ, Du C, Bransby MF, McKenna SJ, Valentine TA: Automated motion estimation of root responses to sucrose in two Arabidopsis thaliana genotypes using confocal microscopy. Planta. 2011, 234: 769-784. 10.1007/s00425-011-1435-7.View ArticlePubMedGoogle Scholar
- Roberts TJ, McKenna SJ, Hans J, Valentine TA, Bengough AG: Part-Based Multi-Frame Registration for Estimation of the Growth Of Cellular Networks in Plant Roots. 18th International Conference on Pattern Recognition. 2006, 2: 378-381. . ICPR 2006. IEEE; 0-0 0Google Scholar
- Rymen B, Coppens F, Dhondt S, Fiorani F, Beemster GTS: Kinematic analysis of cell division and expansion. Methods Mol Biol. 2010, 655: 203-227. 10.1007/978-1-60761-765-5_14.View ArticlePubMedGoogle Scholar
- Schindelin J: Fiji Is Just ImageJ (batteries included). ImageJ User and Developer Conference. 2008, LuxembourgGoogle Scholar
- Abràmoff MD, Magalhães PJ, Ram SJ: Image processing with ImageJ. Biophotonics Int. 2004, 11: 36-42.Google Scholar
- Rosin PL: Unimodal thresholding. Pattern recognition. 2001, 34 (11): 2083-2096. 10.1016/S0031-3203(00)00136-9.View ArticleGoogle Scholar
- Chalana V, Kim Y: A methodology for evaluation of boundary detection algorithms on medical images. IEEE Trans Med Imaging. 1997, 16: 642-652. 10.1109/42.640755.View ArticlePubMedGoogle Scholar
- Williams GW: Comparing the joint agreement of several raters with another rater. Biometrics. 1976, 32: 619-627. 10.2307/2529750.View ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.