UBC Faculty Research and Publications

LAMINA: a tool for rapid quantification of leaf size and shape parameters Bylesjö, Max; Segura, Vincent; Soolanayakanahally, Raju Y; Rae, Anne M; Trygg, Johan; Gustafsson, Petter; Jansson, Stefan; Street, Nathaniel R Jul 22, 2008

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


52383-12870_2008_Article_290.pdf [ 970.94kB ]
JSON: 52383-1.0224011.json
JSON-LD: 52383-1.0224011-ld.json
RDF/XML (Pretty): 52383-1.0224011-rdf.xml
RDF/JSON: 52383-1.0224011-rdf.json
Turtle: 52383-1.0224011-turtle.txt
N-Triples: 52383-1.0224011-rdf-ntriples.txt
Original Record: 52383-1.0224011-source.json
Full Text

Full Text

ralssBioMed CentBMC Plant BiologyOpen AcceSoftwareLAMINA: a tool for rapid quantification of leaf size and shape parametersMax Bylesjö*1, Vincent Segura2, Raju Y Soolanayakanahally3, Anne M Rae2, Johan Trygg1, Petter Gustafsson4, Stefan Jansson4 and Nathaniel R Street4Address: 1Research Group for Chemometrics, Department of Chemistry, Umeå University, SE-901 87 Umeå, Sweden, 2CNAP Artemisia Research Project, Centre for Novel Agricultural Products, Department of Biology, PO Box 373, York, YO10 5YW, UK, 3Department of Forest Sciences, 2424, Main Mall, University of British Columbia, Vancouver, BC, V6T 1Z4, Canada and 4Umeå Plant Science Centre, Department of Plant Physiology, Umeå University, SE-901 87 Umeå, SwedenEmail: Max Bylesjö* - max.bylesjo@chem.umu.se; Vincent Segura - vs523@york.ac.uk; Raju Y Soolanayakanahally - rajus@interchange.ubc.ca; Anne M Rae - amr502@york.ac.uk; Johan Trygg - johan.trygg@chem.umu.se; Petter Gustafsson - petter.gustafsson@plantphys.umu.se; Stefan Jansson - stefan.jansson@plantphys.umu.se; Nathaniel R Street - nathaniel.street@plantphys.umu.se* Corresponding author    AbstractBackground: An increased understanding of leaf area development is important in a number offields: in food and non-food crops, for example short rotation forestry as a biofuels feedstock, leafarea is intricately linked to biomass productivity; in paleontology leaf shape characteristics are usedto reconstruct paleoclimate history. Such fields require measurement of large collections of leaves,with resulting conclusions being highly influenced by the accuracy of the phenotypic measurementprocess.Results: We have developed LAMINA (Leaf shApe deterMINAtion), a new tool for the automatedanalysis of images of leaves. LAMINA has been designed to provide classical indicators of leaf shape(blade dimensions) and size (area), which are typically required for correlation analysis to biomassproductivity, as well as measures that indicate asymmetry in leaf shape, leaf serration traits, andmeasures of herbivory damage (missing leaf area). In order to allow Principal Component Analysis(PCA) to be performed, the location of a chosen number of equally spaced boundary coordinatescan optionally be returned.Conclusion: We demonstrate the use of the software on a set of 500 scanned images, eachcontaining multiple leaves, collected from a common garden experiment containing 116 clones ofPopulus tremula (European trembling aspen) that are being used for association mapping, as well asexamples of leaves from other species. We show that the software provides an efficient andaccurate means of analysing leaf area in large datasets in an automated or semi-automated workflow.Background mental sensing units of plants, and by extensionPublished: 22 July 2008BMC Plant Biology 2008, 8:82 doi:10.1186/1471-2229-8-82Received: 7 May 2008Accepted: 22 July 2008This article is available from: http://www.biomedcentral.com/1471-2229/8/82© 2008 Bylesjö et al; licensee 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.Page 1 of 9(page number not for citation purposes)Leaves are of fundamental importance to plants, repre-senting the power generation facility and aerial environ-ultimately provide the energy for sustaining most terres-trial species on earth. A number of genes known to affectBMC Plant Biology 2008, 8:82 http://www.biomedcentral.com/1471-2229/8/82meristomatic pattern formation (e.g. AS1 and WUS,KNOX and CLV, see [1] for a review of leaf development),the rate of leaf primordia initiation [2] and that contributeto the determination of leaf length (ROT3 [3], LNG [4])and width (AN [5]) have now been identified: less isknown about the determination of leaf size currently.Despite these advances, it remains clear that leaf areadevelopment is a highly complex process that is influ-enced by genetic, hormonal and environmental factors.Quantitative Trait Loci (QTL) mapping of leaf develop-ment and leaf size and shape indicators suggests that thesetraits are under the control of many genes [6-15], with rel-atively few genes identified to date [1]. To advance the cur-rent understanding of leaf area development and finaldimension determination requires the ability to pheno-type large collections of leaves from QTL mapping popu-lations, natural populations and forward genetic screensto identify and quantify loci/mutations influencing leafcharacteristics. As well as being important to the fields ofgenetics, physiology, plant breeding and developmentalbiology, leaf shape parameters are also important as ameans of reconstructing historical paleoclimate condi-tions [16,17], where information on leaf serration (depthand presence/absence) is used to accurately reconstructpast mean annual temperature [18,19]. Leaf size andshape parameters (physiognomy) were initially quanti-fied using gridded paper, where a count of squares wasused to measure leaf size, or through the development ofallometric relationships between length, width and area,with length typically being measured and later used to cal-culate area using a regression model. This approach canwork well within a single species but works poorly whenapplied to mapping populations, where segregation canlead to extensive variation in both leaf area and shapetraits. It is equally inappropriate for forward geneticscreens to identify leaf phenotypes, where the inducedphenotypic changes are unpredictable. For many species,field-portable leaf scanning equipment can be used tomeasure leaf area and blade dimensions. However, suchequipment cannot be used on large leaves and workspoorly on species such as Arabidopsis thaliana due to smallleaf area and the proximity of leaves to the soil. Suchequipment is also often limited in the range of measure-ments provided and, as no digital image is captured, ret-rospective re-analysis using, for example, new softwaretools is not possible. More recently, methods have con-centrated on the capturing of digital images of leaves (orfossils) with subsequent analysis using digital image anal-ysis tools. A number of such tools already exist, but noneof the currently available software was able to fulfil ourneeds. ImageJ [20] is a widely used application for theanalysis of biological images and can be used to analysearea and blade dimensions. However, automated analysiswithin the image. ImageJ offers no method to quantifyleaf serration. The development of tools for measuringleaf area was reported in [21] and [22], however they offerlittle to extend the capabilities of ImageJ. More recently[23] reported the development of LeafAnalyser, which isan excellent tool to facilitate PCA analysis of leaf shapeparameters. However, this tool does not report the type ofdimensions that are typically required by plant breeders,physiologists, geneticists or palaeontologists and the soft-ware was not released as open source, negating the possi-bility of further development by the community. Weadditionally found that the implemented thresholdingfrequently required per-image manual adjustment, mak-ing the automated, rapid analysis of leaves more time con-suming. We were interested in measuring basic leafdimension parameters (area, length, width) as well asmeasures of leaf shape, symmetry, serration number anddepth and the missing area within a leaf (as a measure ofdamage by biting herbivores) in a collection of naturallyoccurring clones of Populus tremula, the Swedish Aspen(SwAsp) collection, that are being grown in common gar-den experiments in the south and north of Sweden [24]and that are being used for association mapping [25]. Thisspecies has well defined, characteristic leaf serrations thatwe had visually observed to show variation betweenclones within the SwAsp collection. We were thereforeinterested to see to what degree leaf serration was undergenetic control. This required a rapid and reproduciblemethod of quantifying leaf size and shape parameter traitsas well as serration characteristics. As was reported in [23],we were also interested to see how well PCA could be usedto describe the variation in leaf area characteristics withinthis collection of trees.ImplementationThe LAMINA software has been implemented in Java as astand-alone graphical application. The software is used toidentify leaf objects and to calculate properties of thoseobjects in an automated or semi-automated fashion.Automated analysis requires no user intervention after set-ting the desired parameters whereas semi-automatedanalysis pauses after each image has been analysed toallow manual adjustment of identified blade dimensioncentre lines (i.e. length and width), which can be impor-tant where leaves are not perpendicular to the imageplane. An example screenshot of the user interface isshown in Figure 1A.Main computational stepsThe computational processes involved can be described inthe following sequential steps.1. Thresholding. As an initial step, global thresholding isPage 2 of 9(page number not for citation purposes)is hard to achieve, as is the simultaneous measurement ofarea and blade dimensions when leaves are not squareperformed to find candidate picture elements (pixels) thatputatively represent leaves. In the thresholding process, allBMC Plant Biology 2008, 8:82 http://www.biomedcentral.com/1471-2229/8/82pixel intensities are reduced from the typical grayscalerange of 0–255 to either 0 (off; pixel is background) or 1(on; pixel potentially belongs to a leaf object). As input,the inverse of the blue channel intensities are used ratherthan the entire RGB image. The rationale behind this strat-egy is that while leaves can be green, orange, red or evenblack, they are very rarely blue. On a white background,non-blue objects can, with high accuracy, be distin-guished from the background using global thresholding.The process of identifying a suitable global threshold canbe performed either manually or automatically. In manualthresholding, the user specifies an arbitrary value t in therange 0–255, where pixels with intensities less than t willbe set to 0 (background) and pixels with intensities equalto or greater than t will be set to 1 (putative leaves). Theautomatic thresholding procedure, on the other hand, triesgenerally suitable for images where objects have fairlywell-behaved shapes, which is true for most leaf objects(see Artemisia annua section for an exception). The auto-mated search procedure can be greedy, in which a localminima is found based on a greedy search starting fromthe mean value of the starting image. Alternatively, theprocedure can be exhaustive, in which the entire range 0–255 is searched for the value t that minimizes the varianceof the thresholded image. The latter procedure is generallymore accurate but also considerably slower.2. Segmentation. Posterior to thresholding, the inputimage has been reduced to a binary image containing pix-els that are either background (0 = off) or potential leafobjects (1 = on). The task of segmentation is to groupnearby pixels into segments (objects) that may potentiallyrepresent leaves. The segmentation starts by assigning anUse of LAMINA to quantify leaf characteristics in the SwAsp collectionFigure 1Use of LAMINA to quantify leaf characteristics in the SwAsp collection. A Screenshot of LAMINA. B Example cropped image generated by LAMINA showing dimension measurements and serration detection. C Example cropped image generated by LAMINA. Cavities (holes) in the leaf lamina are marked in green, serrations are marked in blue and the depth of each serration is marked by a yellow line. Horizontal and vertical centre lines are drawn in red with sub-divisions marked in blue. Boundary coordinates are shown as white circles along the perimeter. D Regression analysis to compare data generated from ImageJ to LAMINA for a set of 50 random images. E Principal Component Analysis loadings plot of X and Y coordinates generated for the SwAsp dataset using LAMINA (50 boundary coordinates per leaf). The leaf in the centre is the value closest to the centre of the cloud and has been oriented to match the distribution of XY values in the loadings plot. Component one appears to represent leaf width (55 % variance) and component two leaf length (27 % variance).Page 3 of 9(page number not for citation purposes)to automatically determine a value of t that minimizes thevariance of the thresholded image [26]. This procedure isarbitrary on pixel as the current segment. The segment isthen iteratively extended with neighbouring, unassignedBMC Plant Biology 2008, 8:82 http://www.biomedcentral.com/1471-2229/8/82on pixels (including diagonal pixels) until no more neigh-bouring on pixels exist. This procedure is repeated until allon pixels have been assigned to an object.3. Filtering. Due to measurement noise and presence ofcontaminants in the image, some objects will not repre-sent actual leaves. To remove dubious objects, filteringcan be performed based on both the area of each object(to remove objects that are too small) and based on thedensity of each object (to remove e.g. black frames sur-rounding the image). The default filtering is non-stringentand will only remove the smallest objects, likely to repre-sent contaminations in the image.4. Object boundaries. The boundary of an object isdefined by the set of on pixels where at least one neigh-bour of each on pixel is an off pixel (i.e. pixels on the sur-face of the object). The identification of the boundarypixels is a straightforward computational process. How-ever, in order to simplify the subsequent steps, the adja-cent boundary pixels are internally arranged sequentially(sorted) within each object. This procedure requires thatdistances are calculated between all boundary pixels andcan be time-consuming for highly irregular surfaces, e.g.Artemisia annua images.5. Cavities. Cavities in the leaf objects can be present dueto e.g. biting herbivore damage, which implies that iden-tification and measurement are of interest. A cavity is bydefinition surrounded by a boundary region that isunconnected to the outer boundary of the object. This dis-tinctive characteristic is used to identify the cavities, seenas 'kinks' in the distances between neighboring boundarypixels. The off pixels that can only be connected to the cav-ity (inner) boundary define the cavity area. In this sense,cavities are defined as missing leaf area (holes) within theleaf lamina and do not account for herbivory starting atthe edge of the leaf, which is computationally more diffi-cult to quantify as it would require retrospective calcula-tion of where the leaf boundary was previously. It isequally hard to distinguish herbivory or wounding at theleaf boundary from serrations. This represents an obviousarea of future extension of LAMINA, but is not a trivialtask.6. Serrations and indents. Starting from a boundarypixel, the longest straight line that can be formed withoutcrossing the object formed by non-boundary pixels issought. The intermediate region between two serrationsdefines an indent. This is implemented in practice by con-necting the starting boundary pixel with boundary pixelsof increasing distances until a non-connectable boundarypixel is found. The last connectable pixel, i.e. one that canusing the latest serration point as the starting pixel. Toallow for small variations in the boundary shape, a con-secutive sequence of k non-connectable pixels are allowedbefore stopping. The parameter k can be adjusted by theuser and determines the overall sensitivity of the serrationidentification algorithm.7. Indent depths. Each indent is surrounded by two serra-tions that can be connected by a straight line. The indentdepth is measured as the longest line to the base of theindent while being perpendicular to the straight line con-necting the surrounding serrations. Due to the discretenature of images, it is not always possible to achieve per-fect perpendicularity, and hence a slight discrepancy inthis angle is allowed.8. Boundary coordinates. From the boundary pixels ofeach object, a fixed number of boundary coordinatepoints can optionally be identified. These are defined asequally spaced points around the surface of the object.The boundary coordinates are normalised against the cen-tre coordinate of the object to make the measurementsindependent of the position of the object in the image.Output from LAMINAAfter processing, LAMINA outputs cropped image filesrepresenting the identified objects after thresholding andsegmentation. This allows the user to have a record of theresults of the image analysis process (Figure 1B showsexample cropped images within the LAMINA user iterfaceand Figure 1C an example generated cropped image). Fur-thermore, a number of quantitative measurements of theleaves are generated. This includes the leaf area, height,width, circularity, number of serrations, indent widthsand depths as well as the boundary coordinates (normal-ised against leaf centre). For parameters that summariseseveral measurements, the output includes the mean,median and standard deviation.Scale calibrationImage measurements do not generally contain any infor-mation regarding the actual size of the image. In order toconvert the pixel-based distances and areas in the leafimage into real quantitative measures, scale calibrationhas to be performed. The aim of the calibration is to deter-mine the actual size of one pixel in millimeters (mm) andis optimally run once, to find the conversion ratio. LAM-INA requires a calibration image to perform this calcula-tion, containing one coloured object (not black) ofknown size on a white background. Ideally this objectshould fill the majority of the image area to maximize theaccuracy of the calibration. After determining the meas-ured pixel size of the image, and by manual input of thePage 4 of 9(page number not for citation purposes)be connected by a straight line without crossing theobject, is the next serration point. The process starts againactual size in mm, the pixel-to-mm ratio can be deter-BMC Plant Biology 2008, 8:82 http://www.biomedcentral.com/1471-2229/8/82mined and employed for all subsequent image calcula-tions.Example applications of LAMINAExploring leaf physiognomy in the SwAsp Populus tremula collectionFull details of the common garden experiment can befound in [24]. Five leaves per replicate of each clone weresampled on a single day in early August 2007 into paperbags and later scanned using a Canon CanoScan 4400 FA4 flatbed scanner at a resolution of 300 dpi. A 40 × 50mm yellow rectangle of card was scanned and used forscale calibration. Images were saved as jpeg files. Themajority of genets (clones) were represented by fourclonal replicates. Our sampling strategy was to select fiverandom leaves from different heights on each replicate aswe wanted to know how plastic (i.e. variable) leaf areawas within and between both genets and intra-genetclonal replicates. The only criterion applied was thatleaves should be mature and should not be from the ter-minal stem, as these leaves are of a fundamentally differ-ent nature in aspens. Images were analysed in LAMINA ina semi-automated work flow to allow for corrections tothe orientation of leaves within the image. Default set-tings were used for all parameters except for the serrationdetection pixel threshold, where 22 was used. The centreline of each leaf was adjusted where required before pro-ceeding to the next image. In total, 412 images containing1879 leaves were analysed, with the LAMINA analysis tak-ing 1 working day (8 h).A random set of 50 leaves were scanned and analysedusing ImageJ [20] and LAMINA. For the ImageJ analysis,images were imported as an image stack, at which pointthey were transformed to 8 bit (greyscale) and thenthresholded using a value of 150 to produce a binaryimage. The trace tool was then used to select each leaf andthe Measure tool used to record the selected area. The scalewas set using the line tool to define a known distanceusing the same calibration image used for the LAMINAanalysis. Data generated were analysed and visualised in R[27]. ANOVA tests were performed using the nlme pack-age to test for clone within population and populationeffects. Principal Components Anlaysis (PCA) was per-formed in SIMCA P (v11.3, Umetrics, Sweden).Benchmarking LAMINA using the complex leaves of Artemisia annuaA. annua leaves are highly complex and we deemed themto serve as a comprehensive test of the ability of LAMINAto extract and reliably quantify leaf area and dimensiontraits. We therefore undertook a more detailed methodcomparison using either glasshouse-grown or field-growngenotypes of A. annua. One mature leaf from six geno-types grown in a glasshouse was used to compare leaf arearonmental, Nebraska, USA) and the same leaves werescanned using an HP Scanjet 3570 c A4 flatbed scanner at300 dpi. A 100 × 1 mm bar was scanned and used for scalecalibration. Three mature leaves from 29 genotypes grownoutside at Stockbridge Technology Centre, Cawood,North Yorkshire, U.K. were scanned using the same scan-ner and used for area analysis in LAMINA and ImageJ.Leaves 20, 21 and 22 (counting down from the top of theplant) were sampled in October 2007.LAMINA analysis was performed using the following set-tings: Manual threshold value of 150, no serration detec-tion.ImageJ analysis was performed by transforming images to8 bit (greyscale) and then thresholded using a value of150 to produce a binary image. A polygon was then drawnaround a leaf and the Analyze Particles tool used to calcu-late the area represented by leaf pixels. The scale was set byscanning a standard ruler and using the line tool to definea known distance. The use of the pixel analysis method,rather than the more automated method used for theaspen leaves, was required due to the complex shape ofthe A. annua leaves. However, this method increases thechance of any noise artifacts in the scanned image beingincluded in the measurement calculations.Testing LAMINA using species with diverse leaf shapesIn order to ensure that LAMINA functioned for a diverserange of species, we sampled leaves of a number of com-mon European tree species as well as various poplar spe-cies and A. thaliana. One to three leaves per species wereanalysed to ensure that leaves were reliably extracted fromthe scanned images. All images were scanned as for theSwAsp trees. Additionally, the jpeg format images used asexample applications in [23] and [21] were downloadedand analysed using LAMINA in order to benchmark oursoftware against these other packages.Results and discussionUsing LAMINA to explore leaf traits in the SwAsp collectionIn order to test LAMINA and to provide us with an over-view of leaf characteristics within the SwAsp collection toguide future experimental design, we sampled leaves fromthe northern common garden of the SwAsp collection[24]. As we had previously used ImageJ [20] for analysingleaf area, we first performed a comparison analysisbetween ImageJ and LAMINA as an initial benchmark toensure that LAMINA provided comparable results. Bothprograms returned effectively identical measures of leafarea (Figure 1D), with an R2 value of 0.99. Having estab-lished that LAMINA was functioning as intended, we thenPage 5 of 9(page number not for citation purposes)meter data to LAMINA. The area of each leaf was meas-ured using a LI-COR LI-3100 Area Meter (LI-COR Envi-extended the analysis to the entire set of sampled leaves.Using this data, we first examined the variation betweenBMC Plant Biology 2008, 8:82 http://www.biomedcentral.com/1471-2229/8/82multiple leaves sampled from the same clonal replicate,which indicated that there was significant variation in leafarea within an individual plant (data not shown). Thisprompted us to extract only the leaf with the largest areameasurement from each replicate, which reduced intra-genet variance, with the results shown in Table 1. TheANOVA model results in Table 1 show that, even afterselecting only the largest leaf per replicate, there was stillsignificant variance in leaf area within a genet. This resultwill be essential to guiding subsequent leaf sample collec-tion and also indicates that very careful considerationshould be given to sample collection not only for mor-phological analysis but also for other purposes such asphysiology, transcriptomics and metabolomics, as leafdevelopment appears to be highly plastic in aspen. Wealso examined the results of a PCA analysis of the bound-ary coordinate data and trait variable data produced byLAMINA. Figure 1E shows the loading plot of XY bound-ary coordinates for the set of data representing the largestleaf from each genet replicate. Both X and Y sets of coor-dinates form spherical distributions but they lie at rightangles to each other. Principal component one appears torepresent leaf width and component two leaf length, withthese two components explaining the majority (82 %) ofthe variance in the data. PCA of the morphological traitvalues showed a distribution pattern confirming the cor-relation results shown in Table 1 (data not shown). Wehave therefore shown that LAMINA is suitable for extract-ing meaningful biological data using different PCAapproaches in a fashion similar to [23] but with the addi-tional advantage that traditional morphological measuresof leaf traits are provided by LAMINA for use in methodsother than PCA. The data produced by LAMINA is equallysuitable for use in other analysis methods.Benchmarking LAMINA against ImageJ and a leaf area meter using Artemisia annuaLeaves from A. annua plants were by far the most complexin structure of those that we used for testing and develop-ing LAMINA. We therefore examined the results generatedfor these leaves in more detail as a means of benchmark-ing LAMINA. Leaf area is an important trait in A. annua asthis medicinal crop produces artemisinin, used in anti-malaria drugs, in glandular trichomes found predomi-nantly on the leaf surface. Natural variation in A. annualeaves is being studied using QTL mapping and associa-tion studies, while induced mutations and phenotypes arebeing identified using forward and reverse genetic screenswith all of these approaches requiring rapid and reliablequantification of leaf area.We performed two small comparisons, one using a leafarea meter and the second using ImageJ as these twomethods of calculating leaf area represent those mostcommonly used currently. Both methods provided effec-tively identical results for leaf area (Figure 2, R2 = 0.9938for the leaf area meter comparison and R2 = 0.9923 for theImageJ comparison). However, LAMINA has the addedadvantage of also providing a suite of additional measure-ments alongside area (although see below), as well as pro-viding a far greater level of automation in the analysispipeline. Both sets of results presented in Figure 2 showthat LAMINA is able to reliably extract leaves fromscanned images and accurately calculate leaf morphologi-cal traits from such complex leaves to a level of accuracythat matches existing, commonly-used analysis methods.LAMINA is suitable for use in a diverse range of speciesTo qualitatively assess the general applicability of LAM-INA, we scanned leaves from a diverse range of tree andflowering plant species. These included species com-monly used in laboratory and genetics/ecology research aswell as a range of species with divergent leaf shapes andforms. A number of examples of the cropped outputimages generated by LAMINA are shown in Figure 3,including the example images of [23] and [21], with theresults showing that LAMINA performs equally well asexisting software tools. We also tested LAMINA on a col-Table 1: Overview of leaf size and shape traits in the SwAsp treesANOVAClone(Pop) Population Latitude LongitudeArea ns ns ns nsLength ns ns ns nsWidth ns ** ns nsLength:Width ns * ns nsCircularity ns ns * nsHorizontal symmetry * ** *** *Vertical symmetry ns ns ns nsNumber of serrations * ** *** ***Indent depth ns ns ns nsIndent width ns * ns *Page 6 of 9(page number not for citation purposes)ANOVA analysis of leaf size and shape parameter data generated using LAMINA. Significance values are * 0.05, ** 0.01, *** 0.001, ns not significant.BMC Plant Biology 2008, 8:82 http://www.biomedcentral.com/1471-2229/8/82lection of scanned images of Populus balsamifera leaves,which have numerous, small serrations. LAMINA was ableto quantify these small serrations provided image resolu-tion was adequate (leaves were scanned at 600 dpi).Current limitations and future developmentWe have shown that LAMINA is able to accurately extractand quantify leaf area from scanned images of a diverserange of plant species. However, there are limitations tothe use of the provided dimension measurements cur-rently provided by LAMINA, and these limitations repre-sent the most immediate targets for future developmentand expansion of LAMINA.Although LAMINA is able to quantify leaf area of the A.annua leaves accurately, there are currently limitations tothe use of additional measurements returned by LAMINA,with these being true in a range of leaf forms. Examiningthe A. annua leaves shown in Figure 3 shows that LAMINAcurrently returns blade dimensions using only straightlines, which is clearly far from ideal in these leaves. It isalso clear that serration and cavity analysis will not returnmeaningful values from these leaves. There is thereforeclear caution and consideration required by end userswhen making use of values returned by LAMINA. In thecase of A. annua, we would suggest that calculated area iscertainly reliable and that circularity may also be a usefulindicator of how that leaf area is distributed. The use ofAs is the case for the A. annua leaves, many leaves do nothave perfectly straight or symmetrical shapes and as such,the central line deviates from a straight line. CurrentlyLAMINA only returns a measure of the maximum (or userset) straight line distance between the leaf base and tip.The inclusion of a tool to additionally allow manualplacement of a non-straight line tracing the centre line(often the central vein) of a leaf is an obvious first targetfor extension of the current measures provided by LAM-INA. At present, the difference between the values of thereturned 25 % and 75 % vertical lengths can be used toindicate leaf asymmetry, which will often reflect thedegree of leaf curvature and therefore the likely inaccuracyof the returned straight line centre measurement.The A. annua leaves and the included example A. thalianaleaves shown in Figure 3 identify another important issueto consider when sampling leaves to be analysed usingLAMINA – that of petioles: If petioles are sampled as wellas actual leaf area, LAMINA will include the petiole as partof the leaf and this will affect generated measurements. Inmany species, removal of petioles is simple as there is aclearly identifiable boundary between leaf and petiole. Ifpetioles are being removed, it is essential that this is doneaccurately as any remaining petiole will lead to the mis-identification of a serration either side of the remainingpetiole. Figure 3G represents a more complex example,but one that is typical for many A. thaliana plants, whereComparison of methods for quantifying leaf area in A. annuaFigure 2Comparison of methods for quantifying leaf area in A. annua. A Comparison of leaf area quantification using a leaf area meter and LAMINA. B Comparison of leaf area data generated using ImageJ and LAMINA.         	      050010001500200025003000350040000 1000 2000 3000 4000   ff fi fl ffiy  = 0.9707x 	       01000200030000 1000 2000 3000 ! " fl ff  fl # fl "$ % & ' ( ( )* +,-../* +,-../$ % & ' ( ( )Page 7 of 9(page number not for citation purposes)any other returned values would require careful consider-ation by the end user.there is no clear boundary between the leaf lamina andthe petiole. In such cases, it is often very hard to defineBMC Plant Biology 2008, 8:82 http://www.biomedcentral.com/1471-2229/8/82where leaf becomes petiole and the sampling strategymust take this into consideration: LAMINA will includeany scanned leaf area when generating dimension meas-urements and the end user must therefore decide whatthey wish to be included at the point of sampling (or bylater manipulation of the generated scanned images toremove e.g. remaining, unwanted petiole area). Althoughit is not inconceivable that an algorithm could be devel-oped to differentiate between leaf and petiole, this is cer-tainly far from a trivial task, especially if such analgorithm should be generally applicable across species.The example A. thaliana leaves in Figure 3G highlightanother point that users must be aware of: currently, thesoftware will not distinguish between wounding at theleaf boundary and serrations. In the cases shown in Figureleaves are frequently curved and can not be flattened with-out tearing the leaf lamina. The use of median rather thanmean serration values will limit the influence of such out-liers but if serration quantification is being used, usersshould visually screen through the cropped images pro-duced by LAMINA to identify problem leaves. It is possi-ble that the algorithm for detecting leaf serrations couldbe extended to differentiate between boundary woundingor grazing herbivore damage (that typically extends fromthe boundary edge into the leaf) and actual serrations.However, as with other similar problems such as petioledetection, this will not be simple if the algorithm is to beapplicable across a wide range of species (for example,many species contain serrations in combination withdeeper, more infrequent, leaf lobes). Such algorithmicdevelopment would require extensive testing and confir-mation across a broad range of species that have beenexposed to a range of herbivore damage and wounding.There are also a number of potential extensions to LAM-INA that we feel would have broad appeal to leaf research-ers, including colour quantification (for example to tracksenescence), detection of necrotic lesions or flecks, meas-urement of leaf rust urediospore number and dimensions,and quantification of veinal pattern. As LAMINA has beenreleased as an open source project using the well-sup-ported Java language, it represents an ideal framework forthe future integration of such extensions by the commu-nity and we hope that the instigation of such an opensource project can serve as a means of concentrating devel-opment of a powerful phenotyping tool, as has been thecase for the analysis of microscopy images since the initialrelease of ImageJ [20].ConclusionWe have developed a new software tool for the automatedor semi-automated analysis of leaf morphological traitsand have shown that the method is able to extract biolog-ically meaningful data from a range of species with con-trasting leaf shapes. The developed software performsequally well as existing software while also providing anextended range of measures of leaf size and shape indica-tors. We show that the software performs as well as com-monly used leaf area meters, even when measuring highlycomplex leaf forms. Application of this software tool willsignificantly aid the rapid screening of large-scale collec-tions of genotypes for forward or reverse genetics as wellas equally serving plant breeders. This is the first opensource tool available for the quantification of leaf serra-tion.Availability and requirementsProject name: LAMINA: Leaf shApe deterMINAtionExample cropped images generated using LAMINA in a range of speciesFigure 3Example cropped images generated using LAMINA in a range of species. A Three example Artemisia annua leaves. Some regions are incorrectly identified as cavities, however the perimeter is correctly identified. B Example image from [21]. Serration detection pixel threshold = 50. C Example image from [23]. D Example Populus leaves from Umeå Plant Science Centre 2006 Calendar. E Example Image containing a range of leaves from common European tree species with contrasting leaf shapes. F Example use of serra-tion detection to measure lobes in a senescing maple leaf. Serration detection pixel threshold = 75. G An example set of Arabidopsis thaliana leaves representing a developmental series. All images were analysed using the Greedy search threshold setting.Page 8 of 9(page number not for citation purposes)3, boundary damage most likely resulted from flatteningthe leaves at the point of image collection, as A. thalianaProject home page: http://sourceforge.net/projects/laminaPublish with BioMed Central   and  every scientist can read your work free of charge"BioMed Central will be the most significant development for disseminating the results of biomedical research in our lifetime."Sir Paul Nurse, Cancer Research UKYour research papers will be:available free of charge to the entire biomedical communitypeer reviewed and published immediately upon acceptancecited in PubMed and archived on PubMed Central BMC Plant Biology 2008, 8:82 http://www.biomedcentral.com/1471-2229/8/82Operating system(s): Platform independentProgramming language: JavaOther requirements: Java 1.4.x or higher. LAMINA usesthe Java Advanced Imaging (JAI) package http://java.sun.com/javase/technologies/desktop/media/jai/ tosupport common image file formats, which is bundledwith the installation and hence no additional installationshould be required.License: GNU GPL2AbbreviationsLAMINA: Leaf shApe deterMINAtion; SwAsp: SwedishAspen; QTL: Quantitative Trait Loci; PCA: Principal Com-ponent Analysis.Authors' contributionsMB developed the LAMINA software and contributed tothe manuscript production. VS performed the A. annualeaf analysis and was supervised by AMR. RYS performedthe SwAsp LAMINA analysis. JT, SJ, PG supervised theproject. NRS conceived the project and tested the soft-ware, drafted the manuscript, scanned the SwAsp leavesand analysed the SwAsp leaf data. All authors read andapproved the manuscript.AcknowledgementsWork on A. annua carried out by VS and AMR is part of the CNAP A. annua Research Project, funded by the Bill and Melinda Gates Foundation.References1. Fleming AJ: The control of leaf development.  New Phytologist2005, 166:9-20.2. Pien S, Wyrzykowska J, Mason S, Smart C, Fleming A: Local expres-sion of expansin induces the entire process of leaf develop-ment and modifies leaf shape.  Proceedings of the National Academyof Sciences, USA 2001, 98:11812-11817.3. Tsuge T, Tsukaya H, Uchimiya H: Two independent and polar-ized processes of cell elongation regulate leaf blade expan-sion in Arabidopsis thaliana (L.) Heynh.  Development 1996,122:1589-1600.4. Lee Y, Kim GT, Kim IJ, Park J, Kwak SS, Choi G, Chung WI:LONGIFOLIA1 and LONGIFOLIA2, two homologous genes,regulate longitudinal cell elongation in Arabidopsis.  Develop-ment 2006, 133:4305-4314.5. Tsukaya H, Tsuge T, Uchimiya H: The cotyledon: A superior sys-tem for studies of leaf development.  Planta 1994, 195:309-312.6. Rae AM, Ferris R, Tallis MJ, Taylor G: Elucidating genomicregions determining enhanced leaf growth and delayedsenescence in elevated CO2.  Plant Cell and Environment 2006,29:1730-1741.7. Street NR, Skogstrom O, Sjodin A, Tucker J, Rodriguez-Acosta M,Nilsson P, Jansson S, Taylor G: The genetics and genomics of thedrought response in Populus.  Plant Journal 2006, 48:321-341.8. Juenger T, Perez-Perez J, Bernal S, Micol J: Quantitative trait locimapping of floral and leaf morphology traits in Arabidopsisthaliana: evidence for modular genetic architecture.  Evolutionand Development 2005, 7:259-271.9. Wullschleger S, Yin T, Difazio S, Tschaplinski T, Gunter L, Davis M,10. El-Lithy M, Clerkx E, Ruys G, Koornneef M, Vreugdenhil D: Quanti-tative Trait Locus Analysis of Growth-Related Traits in aNew Arabidopsis Recombinant Inbred Population.  Plant Phys-iology 2004, 135:444-458.11. Frary A, Fritz L, Tanksley S: A comparative study of the geneticbases of natural variation in tomato leaf, sepal, and petalmorphology.  Theoretical and Applied Genetics 2004, 109:523-533.12. Perez-Perez J, Serrano-Cartagena J, Micol J: Genetic Analysis ofNatural Variations in the Architecture of Arabidopsis thalianaVegetative Leaves.  Genetics 2002, 162:893-915.13. Jiang C, Wright RJ, Woo SS, Delmonte TA, Paterson AH: QTL anal-ysis of leaf morphology in tetraploid Gossypium (cotton).Theoretical and Applied Genetics 2000, 100:409-418.14. Wu R, Stettler R: Quantitative genetics of growth and develop-ment in Populus. III. Phenotypic plasticity of crown structureand function.  Heredity 1998, 81:299-310 [http://www.nature.com/hdy/journal/v81/n3/abs/6883970a.html].15. Wu R, Bradshaw HD, Stettler RF: Molecular genetics of growthand development in Populus (Salicaceae) .5. Mapping quan-titative trait loci affecting leaf variation.  American Journal of Bot-any 1997, 84:143-153.16. Wilf P, Wing SL, Greenwood DR, Greenwood CL: Using fossilleaves as paleoprecipitation indicators; an Eocene example.Geology 1998, 26:203-206.17. Krieger JD, Guralnick RP, Smith DM: Generating empiricallydetermined, continuous measures of leaf shape for paleocli-mate reconstruction.  PALAIOS 2007, 22:212-219.18. Huff PM, Wilf P, Azumah EJ: Digital Future for Paleoclimate Esti-mation from Fossil Leaves? Preliminary Results.  PALAIOS2003, 18:266-274.19. Royer DL, Wilf P, Janesko DA, Kowalski EA, Dilcher DL: Correla-tions of climate and plant ecology to leaf size and shape:potential proxies for the fossil record.  Am J Bot 2005,92:1141-1151.20. ImageJ   [http://rsb.info.nih.gov/ij/]21. Igathinathane C, Prakash VSS, Padma U, Babu RG, Womac AR: Inter-active computer software development for leaf area meas-urement.  Computers and Electronics in Agriculture 2006, 51:1-16.22. Bakr E: A new software for measuring leaf area, and areadamaged by Tetranychus urticae Koch.  Journal of Applied Entomol-ogy 2005, 129:173-175.23. Weight , Caroline , Parnham , Daniel , Waites , Richard : LeafAna-lyser: a computational method for rapid and large-scale anal-yses of leaf shape variation.  The Plant Journal 2008, 53:578-586.24. Luquez V, Hall D, Albrectsen B, Karlsson J, Ingvarsson P, Jansson S:Natural phenological variation in aspen (Populus tremula):the SwAsp collection.  Tree Genetics & Genomes 2007.25. Ingvarsson PKK, Garcia V, Luquez V, Hall D, Jansson S: Nucleotidepolymorphism and phenotypic associations within andaround the phytochrome B2 locus in European aspen (Populustremula, Salicaceae).  Genetics 2008.26. Gonzalez RC, Woods RE: Digital image processing 2nd edition. Pren-tice Hall; 2002. 27. Ihaka R, Gentleman R: R: a language for data analysis and graph-ics.  Journal of Computational and Graphical Statistics 1996, 5:299-314.yours — you keep the copyrightSubmit your manuscript here:http://www.biomedcentral.com/info/publishing_adv.aspBioMedcentralPage 9 of 9(page number not for citation purposes)Tuskan G: Phenotypic variation in growth and biomass distri-bution for two advanced-generation pedigrees of hybrid pop-lar.  Canadian Journal of Forest Research 2005, 35:1779-1789.


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items